- articleHow do reverberation tail length, binaural versus diotic reverberation and early reflections in a headphone render affect externalization and localization, and what do open-access sources say about how much room detail to include?
Two open-access reviews (Best et al. 2020; Neidhardt et al. 2022) report, from cited primary studies not re-read here, that externalization improves with the first roughly 80 ms of binaural reverberant tail, depends on interaural differences in the reverberation, and matters more for midline sources whose direct sound carries weak binaural cues. Neidhardt et al. report that sensitivity to early-reflection detail is limited, but simplified early reflections can still alter coloration, apparent source width and localization. No primary open-access full text that varies RT60, DRR or early-reflection patterns on a headphone render was retrieved; the quantitative claims are therefore review-reported.
- articleWhy does binaural headphone audio often sound inside the head, and which cues does a spatial-audio workflow need to provide so a listener hears it outside the head?
A 2020 open-access review (Best et al., Trends in Hearing) reports that diotic headphone sound is typically heard inside the head and that externalization is more likely with binaural reverberation, with sources that stay fixed in the world when the head turns, and with cues consistent with what the listener expects of the room. Evidence on whether listener-specific HRTFs are needed is described as mixed. The review is a secondary, keyword-search synthesis; its statements are leads to the cited primary studies, of which one (Gil-Carvajal 2016) was read in full here and others are covered by catalogue ips012.
- articleWhat defects or hardware/format limits for Resolve Voice Isolation do Blackmagic's own update notes mention, and which versions do they belong to?
Blackmagic's support-listing notes mention three Voice Isolation fixes, all in Studio builds: short audio clips (18.1.1, 23 Nov 2022), playback on energy-efficient CPU/GPU systems (18.5.1, 14 Aug 2023) and audio sync when bouncing (18.6.3, 14 Nov 2023). The notes say 'improves' and do not describe the earlier behaviour or give thresholds, so what exactly went wrong in earlier builds is unknown. No later (19, 20, 21) update note in the listing mentions voice isolation.
- articleWhen a split edit moves AAF to OTIO (and on to EDL, FCP7 XML or FCPXML), what survives as timing and what is lost as pairing between the audio and video clips, according to specs, adapter code and the OTIO linking issue?
AAF and OTIO both place audio and video on separate tracks, so a split edit is just two tracks whose clip boundaries differ; the AAF Edit Protocol calls such tracks synchronized but independent. Neither OTIO nor its AAF, CMX3600 and FCP7 adapters (as read at pinned commits) has a construct for saying the two clips belong together: OTIO issue 343 has been open since 2018, the CMX adapter TODO admits linkage is lost, and the FCP7 adapter drops link elements on write. Users report Resolve and Premiere put link ids in OTIO metadata, which is unverified. Net by reading: timing offsets can travel, pairing does not.
- articleHow does FCPXML represent a split edit (audioStart / audioDuration), on which elements, in which units, and does the OTIO FCPX adapter read it?
Apple's FCPXML DTD (1.7 archive and the 1.10 page) declares optional audioStart and audioDuration time attributes on clip, mc-clip, ref-clip, sync-clip and asset-clip and says they define J/L cuts (split edits) on composite A/V clips; times are rational seconds. Two FCPXML 1.11 files in a third-party Swift library contain the attribute pair on sync-clips, with audioStart equal to start and audioDuration shorter than duration. The OTIO FCPX adapter at c61839e1 never mentions either attribute and skips clip/sync-clip/mc-clip as items, so a split edit expressed this way is not read by it. The sign convention of audioStart relative to start and offset is not defined in the DTD text; a third-party importer treats any difference between audioStart/audioDuration and start/duration as a split edit and then follows the video timing.
- articleWhen a CMX3600 EDL with a SPLIT: line is read by the OTIO CMX3600 adapter or by pycmx, what happens to the delay value, and what is lost when OTIO writes an EDL back?
By code reading (nothing executed): four open-source EDL readers all fail to apply the SPLIT delay. The OTIO CMX3600 adapter at 4b02a666 recognises a SPLIT line, checks that exactly one delay is named, then consumes the next two event lines and discards the delay value; the per-track start offset may still differ between the video and audio tracks because first-clip placement is stored in track source_range. pycmx at 49a536f0 parses SPLIT: into a StmtSplitEdit that its Event/Edit classes never expose, and the statement is stored with the previous event. The OTIO writer cannot emit SPLIT, AUD or combined V+A channel codes and writes only the first track. A Rust OTIO port documents that the delay is not modelled because the row times already carry it, and a web editor's importer warns the user that it is not applied. Applied by hand to a real single-spaced SPLIT line, pycmx's fixed columns would mis-read both type and delay. None of the inspected test files contains a split-edit case.
- articleI need to hand a J-cut or L-cut from one editing tool to another: which interchange format has a field for it, and what do the OTIO adapters do with that field?
Each format expresses a split edit differently. A CMX3600-style EDL uses a SPLIT: header plus paired video and audio rows (one worked example). FCPXML names audioStart and audioDuration as the J/L-cut mechanism in Apple's DTD comment. AAF has no split object: audio and video are separate slots that its edit protocol calls synchronized but independent. OTIO has separate tracks and no clip-linking construct (its linking issue is open). Of the open-source EDL readers inspected, the OTIO adapter validates the split marker and discards its value, a Rust OTIO port documents not modelling it, and one importer warns that it is not applied; the FCPX adapter never reads audioStart/audioDuration. Treat all of this as code-reading, not execution.
- articleWhat does a split edit (audio and video edit points at different times) look like in a CMX3600-style EDL, and how do I read which track is delayed and by how much?
A split edit in a CMX3600-style EDL is a "SPLIT: AUDIO DELAY= <timecode>" (or VIDEO DELAY) header followed by two rows with one event number, a picture row and a sound row. A 1995 vendor guide example (audio 6 frames late, video-only row plus A row) and a real Avid-style list found in a public repository (video 6 frames late in one block, audio 1 s 19 f late in another) agree on the structure: the delayed row is listed second and the header value equals the difference of the rows' record-in values (at 24 fps for the real list). A practitioner explainer says the header states which element is delayed and by how much. Header spacing differs between the two examples. No formal CMX or SMPTE field definition was read, so this rests on two examples, one explainer and parser code.
- articleWhich objective metrics (PESQ, STOI, DNSMOS, SI-SDR) and listening tests do the DeepFilterNet papers and the metric papers report, what do those numbers measure, and how far can they stand in for post-production acceptance of cleaned dialogue?
The DeepFilterNet papers report PESQ, CSIG, CBAK, COVL and STOI on Voicebank+Demand and, for DFN2, DNSMOS P.835 on the DNS4 blind set; neither the DFN2 nor the DFN3 paper (v1 HTML read) reports a listening test. DNSMOS P.835 is a non-intrusive predictor trained on crowd P.835 ratings of suppressor output, accurate per model (PCC 0.94-0.98) but weaker per clip (0.71-0.83). Intrusive metrics need a clean reference, which production dialogue lacks; SI-SDR ignores gain error. In URGENT 2024 two systems topped DNSMOS/NISQA yet scored low on MOS. No source read links any of these metrics to a post-production acceptance threshold.
- articleI only have the built-in enhance function (Premiere Enhance Speech, Adobe Podcast Enhance Speech or Descript Studio Sound): what are its documented limits and pitfalls, and what should I know before using it on film dialogue?
Adobe and Descript help pages state concrete limits: Premiere Enhance Speech takes mono or stereo and returns a mono downmix, is clip-based, is documented as running on-device, may leave "unexpected or unprocessed words or syllables" at very low SNR, struggles with beamforming-mic recordings, and needs Media Encoder add-ons for matching exports. Descript Studio Sound needs internet, applies per file, is for recorded (not synthetic) speech, can output a silent waveform on very loud noise (lower Intensity) and has a documented Chrome export drift bug. Adobe Podcast Enhance Speech returns the uploaded format, with a 1 GB cap on the capture read. None of these pages publishes a measurement.
- articleFor iZotope RX Dialogue Isolate, Repair Assistant and Scene Rebalance, which controls, quality modes and documented failure modes should a dialogue editor know, and what changed between the RX 11 and RX 12 documentation?
iZotope documents Dialogue Isolate as an ML separator with three gain controls, a Sensitivity control that it says trades noise removal against artifacts and dialogue clarity, a Good/Real-time vs Best/Offline mode (Advanced), and four-band scaling. Scene Rebalance adds a Voice/Music/Effects split whose voice Sensitivity keeps more breaths but can leak other elements. Release notes record an offline-mode defect (muffled end of selection) fixed in 11.2.0. The one first-person review read reports artifacts when the noise control is pushed and a disappointing Repair Assistant result. No source here measures quality.
- articleFaced with a noisy production-dialogue clip, what do the common AI "enhance" and "isolate" tools (iZotope RX, Premiere Enhance Speech, Adobe Podcast, Descript Studio Sound, DaVinci Resolve Voice Isolation) take in, give back and let me control, per their own documentation?
Vendor documentation describes two families. One-control tools return a single processed dialogue track with one blend or intensity control (Premiere Enhance Speech Mix Amount, Descript Studio Sound Intensity; the Adobe Podcast pages read describe no strength control). Component tools split the input and expose per-component gain (iZotope Dialogue Isolate: voice, reverb, noise; RX 12 Scene Rebalance: voice, music, effects). The Resolve 20 guide read shows Voice Isolation as a toggle (Voiceover tool, AI Audio Assistant) but not its parameters; the 209 MB reference manual was not fetched. All statements below are vendor statements, not tests.
- articleWho keeps the consent evidence and the cloned-voice data in ElevenLabs' and Microsoft's texts, and what can the speaker or customer delete?
ElevenLabs' Creative Terms put consent records on the customer (five years or the legal period, shown on request); the Terms let the user request deletion of User Voice Models and opt out of training use of Content, with effect only going forward. Microsoft's Product Terms let Microsoft retain a copy of each Synthetic Voice and acknowledgement audio under a customer-granted licence; Microsoft's talent disclosure says it keeps the recorded statement as long as needed for security and integrity and may keep a copy of custom neural voice models; the Code of Conduct requires apps serving external users to let them remove their voice. Texts compared at dated versions; a synthesis of vendor statements, not a legal analysis.
- articleWhen I need to clone or convert a real person's voice with a hosted provider, what consent proof does each provider ask for, in the first place?
As observed 2026-10-22 the providers read ask for three different things. Microsoft (personal and professional voice), Google (Chirp 3 Instant Custom Voice) and OpenAI (recording-based custom voices) each require a separate recorded consent statement, matched to the speaker of the sample, and gate access to approved customers. Microsoft and Google prescribe the wording; OpenAI has the caller read one of its listed phrases. ElevenLabs asks for a confirmation (instant clone) or a recorded verification of your own voice (professional clone). Resemble's terms say consent may be required and must be verbal if so, but its voice-cloning pages are now archived historical documentation with no consent step.
- articleGiven what the open evaluation tools measure, how should an agent evaluate a TTS or voice-conversion take for narration or dubbing?
This candidate is a synthesis (a proposal, not an executed result) built from candidates 01 and 02. No source read defines a dubbing or narration acceptance test. The sources support a layered check: an ASR round-trip for intelligibility, a speaker-embedding cosine against the actual reference audio for voice identity, and a small blind listening comparison for naturalness, expressiveness and timing, with counts and the comparator reported. Timing, lip-sync, emotion and artefacts are not covered by any tool read here and need listening.
- articleHow do two open zero-shot TTS papers run their listening tests (CMOS and SMOS), how many raters do they say they used, and what is missing for judging whether a score difference is real?
Seed-TTS (arXiv v1) uses a comparative MOS where raters hear a reference clip then the system and a human recording in random order, scores -2 to +2, and calls differences under 0.1 insignificant, but states no rater count or interval for it and does not release the subjective set. F5-TTS (arXiv v3) states 20 native raters per language over 30 rounds on a -3 to +3 CMOS plus a 1-5 SMOS, and itself lists other papers' rater counts (6, 12). In the extracted text of both papers there is no confidence interval, standard error or inter-rater statistic, so differences such as Seed-TTS CMOS -0.07 versus -0.08 cannot be assessed from the reports.
- articleWhat minimum text sizes or proportions do broadcaster and streamer documents set for on-screen text in video, and are any given as x-heights?
DPP/Channel 4 state a minimum SD font height of 20 SD lines and derive 40 HD and 80 UHD lines/pixels for burned-in text so it stays legible after down-conversion (arithmetic: 3.7% of frame height). BBC Subtitle Guidelines set an authoring line height of 7-8% of active video height (4% for 9:16) and let players present at 0.6-0.8x for most screens, never larger than authored. Netflix's Titles & Graphics page states no size minimum. No source read states an x-height or a contrast ratio for video text. Apple's tvOS 23 pt minimum is a UI guideline, not a video-graphics rule.
- articleHow do production sound crews wire actors efficiently and keep them comfortable on set?
Antoin Cox (DPA-published) advises wiring in the trailers or village once the cast is fully in costume, adding a fourth crew member for wiring, and letting actors who dislike being touched be miked with people around them; after a break fix the rig first and play back to show it works. Tanenbaum and Mooney add that daily wiring is made routine early in a shoot and that actors used to radio mics help the sound department. This is practitioner etiquette and logistics, not safety guidance; it does not replace production policy on touching, intimacy or closed sets, which another cycle-3 topic node covers.
- articleWhen a mixer records a boom and lavs of the same speaker, what must be captured and avoided on set so post can combine them (alignment, comb filtering, matching)?
Sources describe a workflow in which production records every mic on its own iso track, sets input gain carefully, and logs the mix; the dialogue editor rebuilds the mix, uses lav isos to avoid perspective jumps between scenes, and aligns boom and lav in time to avoid comb filtering. DPA's vendor article gives rules for when AI dialogue matching works (no more reverb than the reference, no compression or distortion in the source), and DPA's comb-filtering article states the within-10-dB condition. The sources are practitioner statements and vendor demonstrations; none reports a controlled measurement on film dialogue.
- articleFrom the production mixer's side, when should room tone and wild lines be recorded and how do they reduce ADR?
Two named mixers give on-set rules: record room tone and wild lines early with set lighting on and with the mic in the master position (Jim Tanenbaum, Local 695 magazine, copyright 2019); log the mic and its position for each shot to help ADR matching; and on a show that avoids ADR, record multiple wild takes of troublesome lines on location (Stephen Tibbo, Modern Family, CAS Quarterly Spring 2015, where the post mixer reports about one or two ADR lines per episode). Neither source gives a duration for room tone; the 30-60 second figure in the earlier cycle-3 room-tone candidate comes from other sources. Tanenbaum's column pairs these tips with advice to mislead directors and actors, which this candidate records as his tactic and not as a method to follow.
- articleHow do production mixers hide lavalier microphones in wardrobe while keeping clothing noise and tonal loss under control?
Named practitioners treat wardrobe as a department-level problem: wire after costume is final, prepare the wire in the costume, work with costume for transmitter pockets, cable routes and quieter fabrics, and carry an improvised mounting kit. A post engineer's single-speaker test ranks cotton best and silk and polyester coat worst, and DPA's own measurement shows chest and neck placements lose consonant-range high frequencies, so placement under clothing needs EQ. No step-by-step manufacturer note on taping or moleskin could be pinned (Audio-Technica returned 403; Rode, Sennheiser and Lectrosonics pages were not found).
- articleWhat does a production sound report record, and what do mixers and recorder makers say it is for?
In the pinned Sound Devices documents a sound report is a per-folder, per-take log generated from recorder metadata: project header, date, rates and timecode mode, then rows with file, scene, take, length, start timecode, wild-track and circled flags, free-text notes and the name given to each track. A named production mixer (Simon Hayes) says dialogue editors read the reports to see which iso tracks fed the mix, and another (Jim Tanenbaum) advises logging each shot's mic choice and position for ADR. The field list is one vendor tool's default, not an industry standard; the BWF/iXML metadata layer was already covered in cycle 2 and is not repeated.
- articleWhat do Avid's public documents state about Media Composer's AI-assisted dialogue tools (PhraseFind AI, ScriptSync AI, Transcript Tool, Multicam Auto Cut): what they produce, licensing prerequisites, languages and limits?
Avid's Media Composer 2026.8 What's New (31 Aug 2026) documents PhraseFind Multicam Auto Cut (camera switching from speaker assignments in PhraseFind AI transcripts) and Transcript Tool versioning (the original AI transcript is kept and edits stored separately). The 2026.8 ReadMe says the Transcript Tool needs an active PhraseFind AI option (ScriptSync AI alone excludes it) and that indexing runs in the background and is resource intensive. The only language statement found is the 2023 press release (21 languages for automatic detection, initial). No inspected Avid document says whether transcription is on-device or cloud; the 43 MB Editing Guide could not be fetched under the 25 MB rule.
- articleWhat do Blackmagic's guides state about DaVinci Resolve's transcription-driven editing features (text-based editing, speaker detection, IntelliScript, Multicam SmartSwitch): inputs, outputs, edition, matching rules and limits?
Across the Resolve 19, 20 and 21 New Features Guides, transcription underpins text-based editing in the Transcription window (Resolve 19, Studio), speaker detection, SRT import/export, and IntelliScript (Resolve 20, Studio) which builds a timeline from a plain-text script and picks the best take by spoken-dialogue matching only. Resolve 21 adds Final Draft (.fdx) script import. Multicam SmartSwitch (Resolve 20, Studio) cuts to the active speaker using audio plus video cues. The guides inspected do not state language lists for transcription (they point to Project Settings and the reference manual) or whether transcription is local or cloud; the 209 MB reference manual was not fetched.
- articleWhat does Adobe say Premiere's Text-Based Editing takes as input, what it produces, which languages it supports and what its stated limits are?
At the Adobe help-page versions pinned here (pages dated Jan 2026 to Jun 2026), Text-Based Editing auto-transcribes spoken dialogue into a timecode-aligned transcript and mirrors cut/copy/paste of transcript text onto the timeline. Adobe lists UI languages on the overview page and a separate 18-language Speech-to-Text list; the two lists differ, and no inspected page says whether transcription runs on-device or in the cloud. Stated limits are short: dialogue only, no captions workflow inside the feature, clean audio recommended.
- articleWhat are the building blocks of a film soundtrack (dialogue/ADR, effects, backgrounds, Foley, walla, music) and what does each department actually produce before the mix?
Two first-person-heavy sources (Weis 1995 with practitioner interviews; Thom 2000 interview) and two short glossary/textbook excerpts give a working vocabulary: dialogue and ADR are cleaned or re-performed production speech; effects tracks are built from scratch from library, new recordings or combinations; Foley is effects performed to picture (footsteps, cloth, props); walla is a prepared, improvised crowd murmur; a typical final mix groups these as separate elements. The term 'hard effects' does not occur in any page read, so this batch gives no source-based definition of hard effects versus backgrounds beyond the separate 'backgrounds' element in Weis's track layout.
- articleHow does a professional rakugo performer say pauses (ma), first lines, word repetition and breath create understanding when there are no sets, costumes or narration?
Katsura Utanosuke, a professional Kamigata rakugoka, explains in a 2011 talk (printed in a university repository) that rakugo has no scenery, costume or movement, so the audience must imagine 'what is not there'; the only means is speech technique. He defines ma as the blank time where no line is spoken and gives concrete rules illustrated on a short beginners' piece, 'Kuchinashi': set the tone with the first line, show two characters by contrasting pitch, vary repeated words to avoid boredom, pause before a keyword and hold the breath, pause again after a strange phrase so the audience absorbs it, and repeat an identical line but deliver it differently.
- articleWhat is a rakugo performance made of (makura, main story, ochi), and what does one seated performer with a fan and hand towel actually do in it?
Rakugo is a seated solo comic or dramatic story performance in which one performer in kimono plays every character with only a folding fan (sensu) and a hand towel (tenugui). An official Agency for Cultural Affairs introduction and a 2019 speech-synthesis paper agree on the arc: an audience-warming makura (preliminary talk, often improvised), the main story (almost all character dialogue), then a closing line, the ochi (also called sage). The paper adds that Edo-style rakugo can also have an optional opening greeting (maeoki) and a conclusion (musubi) used instead of ochi in some stories. These are Edo/Tokyo-centred descriptions; the Kamigata (Osaka-Kyoto) tradition adds a desk and clapper (see -02).
- articleHow do the audio-capable world-model papers measure that sound stays in sync with actions, camera motion and events, and what do they leave unmeasured?
Three different measurement designs exist. AVWM compares the audio-energy vs motion-energy cross-correlation lag and peak of generated audio with that of ground-truth audio against the true visual trajectory (simulator data, 0.15 s steps); HelixBench (1,015 clips) scores DeSync on an Event subset and a Grounding-DINO-based three-region stereo score for camera-aligned panning; MaineCoon reports an unexplained "AV-Al" column where it is behind Ovi. Each protocol scores windows or short rollouts; none of the texts read measures the delay from a user action to the sound onset.
- articleWhat do open sources say about gain staging and headroom for dialogue, and what do they say (and not say) about EQ for dialogue clarity?
Gain staging and headroom are covered qualitatively: Ardour says meters are for avoiding overload and keeping headroom when tracking, defines 0 dBFS as a system limit with no musical meaning, and offers K20/K14/K12 meters with reference scales at -20/-14/-12 dBFS; Netflix's dubbing guidance asks for no compression, limiting or EQ during recording. For EQ, the REAPER guide only says to use ReaEQ gently, and its 80 Hz / 6 kHz examples are detector filters on a compressor, not corrective EQ. No inspected primary source gave dialogue EQ band frequencies or boosts; that gap is recorded as a negative result.
- articleWhat is room tone, how long should I record it, where do I use it, and what goes wrong if I skip it or process it badly?
Room tone is the ambient sound of a space recorded with everyone silent. A practitioner post (iZotope, 2021) and a trade-press guide (No Film School, 2025) agree on 30 to 60 seconds, with the same microphone placement as dialogue, extended when the ambience changes; its uses are filling gaps between dialogue edits and layering under ADR or overdubs recorded elsewhere. Audacity's manual adds a numeric floor for noise profiling (2048 samples, longer is better) and warns that heavy noise reduction damages the remaining audio. The 30-60 s figure comes from secondary/vendor sources, not a standards body.
- articleWhy does a compressor distort, and which attack, lookahead and release choices does a manual say reduce overshoot and distortion (and when should I use a limiter instead)?
Audacity's development manual explains with reported test-tone measurements that compression is waveshape modification, so it distorts; zero smoothing hard-clips; attack lets the signal overshoot; distortion spreads to every component in the mix; and lookahead/release only help when longer than the period of the lowest-frequency component above threshold. Its examples (1 kHz tone with bursts, -20 dB threshold) show lookahead 10 ms / attack 1 ms / release 10 ms reducing overshoot, and limiter-like 1 ms lookahead / 0 attack / 25 ms release allowing 10 dB more level without clipping. The scope is synthetic tones, not speech.
- articleWhat does a music editor deliver to the dub stage (track grouping, cue sheet, sync check), and what happens to the music there?
The same TV music editor (S5, S6) integrates the recorded cues into the editing session, checks sync, sends the session to the dub stage and prepares a hand-drawn-style dubbing cue sheet "road map". Cues are grouped onto MUSIC-A to MUSIC-D track groups so overlapping cues sit on separate faders (34 music faders in his example). At the dub, dialogue is described as king, music is fitted around it, producers take notes in a final playback and the editors make fixes; mixers then make the 5.1 and stereo masters, an M&E and a DME master. Music-editor deliverables and dub practice here are from one US animated series in 2011.
- articleWhat must be prepared for the recording session itself: click and count-off choices, recording order, tempo notation, and what composers hand to orchestrators?
A music editor (S4, S5) checks that fixed elements sit at the right timecode in a Pro Tools template and records cues in an order driven by player count under a three-hour union minimum. A studio conductor (S9, S11) decides free-click counts (four below 85 bpm, eight above), asks for eighth-note clicks at slow tempos with motor rhythms, marks tempo changes as relative bpm and treats the printed tempo as a guide because the Pro Tools tempo map and timecode are the official references. An orchestrator (S10) lists composer deliverables as MIDI file, full mix and stems, and sends back a new tempo map if tempos are retimed.
- articleHow could an agent mark J-cuts and L-cuts in an edit plan so that a human or tool can build and check them?
No inspected source defines a notation for marking split edits in a plan. The pieces exist, though: Salt measures every dialogue cut by two frame counts (A and B) with L-edits as negative A; Apple and Adobe describe the edit as separate audio and video edit points on one clip, with media handles needed. This candidate proposes a small record per cut built from those pieces. It is a synthesis, untested.
- articleWhat are J-cuts, L-cuts and split edits, which direction is which, and what must be in place to make one in an editor?
A split edit makes the sound and picture cut at different times. A J-cut lets the next scene's sound start before its picture; an L-cut lets the previous scene's sound continue over the next picture. Apple's Final Cut Pro guide (page Last-Modified 2026-09-29) and Adobe's Premiere help (page text last updated 2025-08-22, read via an Internet Archive snapshot) agree on the definition and both give the same prerequisite idea: unlink or expand audio from video and move one edge. An open Ohio State textbook (Copyright 2024) gives the same definitions with film examples.
- articleHow do pansori jangdan (rhythmic cycles) and jo (modes) correspond to mood and scene function, how reliable is that mapping, and what pacing technique can an audio or screen storyteller take from it without pretending to have a pansori drummer?
Sources describe a toolkit of rhythmic cycles from very slow (jinyang, 24 beats) to urgent and agitated (jajinmori) and cheerful (jungjungmori), chosen to follow mood and context, plus modes (ujo, pyeongjo, gyemyeonjo) defined by sentiment and ornament rather than by pitch set alone. The translated music-theory article shows the mapping from cycle to mode is conventional but contested: the same scene has been labelled ujo or pyeongjo depending on the theorist, and slow and medium cycles mix ujo and pyeongjo features. A screen or audio storyteller can borrow the idea of a deliberate tempo-and-mood plan per scene while treating the specific mappings as one tradition's conventions.
- articleOn a scripted set, how do experienced mixers and post people decide between boom and lavalier microphones for dialogue, and what does each choice cost?
In the pinned practitioner accounts the boom is described as the more natural-sounding, perspective-preserving option but one that needs frame headroom, light placement and a skilled operator; lavs are the fallback when wide or multi-camera framing, boom shadows or a noisy room block the boom, at the price of clothing rustle, muffled tone and flat perspective. Working mixers in the CAS Quarterly and Local 695 pieces often record both: boom under the lavs in the mix and every mic on its own iso track so post can choose. These are stated preferences by named people, not measurements.
- articleBefore creating an Azure personal voice or custom (professional) voice, what consent must the speaker give, which Microsoft page says it, and which of those pages is the Product Terms text?
Microsoft's pages use different words for consent. The personal voice overview (ms.date 2026-09-09) requires explicit consent and a recorded statement from the user. The custom voice overview (ms.date 2026-09-04) requires a recording of the voice talent's consent statement before fine-tuning. The Limited Access page and the talent disclosure (both ms.date 2026-03-31) say customers warrant explicit written permission from voice talent. The Azure Product Terms (Internet Archive copies of 2025-10-10, 2026-02-26 and 2026-10-06) put the same written-permission warranty in a paragraph headed Customized TTS Services and use of Synthetic Voices, and let Microsoft require acknowledgement audio. The Product Terms list custom neural voice (later 'custom voice') as a Limited Access Service; the text read does not name personal voice, while the Limited Access Learn page does.
- articleDo the EEA/Switzerland/UK Terms treat voice or biometric Input differently, and how does that sit next to the voice-ownership check described in ElevenLabs' Privacy Policy?
Only the EEA Terms' 4(h) bars Input containing biometric data 'for the purpose of uniquely identifying a natural person', 'except as permitted by applicable law'. The Privacy Policy (Updated 9 October 2026) says voice cloning may ask users to record themselves to confirm the voice is theirs, that this may be biometric or special-category data under some laws, and that it relies on consent for it. The two documents are not reconciled in the text read; both Terms versions also point to a 'Voice Processing Notice in our Privacy Policy', and no heading with that name was found in the Privacy Policy text. The Privacy Policy names ElevenLabs Poland sp. z o.o. as data controller for EEA/UK/Switzerland residents, while the Terms are with Eleven Labs Inc. Presented for curation; no legal conclusion.
- articleDid the Azure Product Terms' text on synthetic voices change between October 2025 and October 2026 copies, and what additional consent, opt-out and disclosure duties does the Microsoft AI Services Code of Conduct put on apps that let external users create voices?
In three Internet Archive copies of the Azure Product Terms (2025-10-10, 2026-02-26, 2026-10-06) the Permissions and Permitted uses paragraphs on synthetic voices carry the same obligations, with wording differences limited to 'text to speech' becoming 'TTS' and 'all TTS implementations' becoming 'all implementations of TTS Services'; the Limited Access list renames 'custom neural voice' to 'custom voice' in the 2026-10-06 copy. The Code of Conduct for Microsoft AI Services (version 4.0, 5/1/2026) separately requires applications offering voice features to external users to allow opt-out and removal of the voice, to forbid pre-existing recordings as training data, to limit each user to their own voice, to obtain verbal consent through a Microsoft acknowledgement statement, and to disclose the synthetic nature of generated voices. Observations only; the live Product Terms could not be fetched.
- articleWhere do Microsoft's short summaries of personal/custom voice consent differ from its Limited Access page, disclosure document and transparency note?
The personal-voice overview summarises the requirement as a speaker's verbal statement. The Limited Access page and the talent-disclosure document add contractual layers: written permission warranted by the customer, advance sharing of a disclosure document, Microsoft's right to run speaker recognition on the consent clip against the training audio, retention of the clip and the model, mandatory disclosure of synthetic nature to users, and use-case approval. The transparency note adds a bundled video-translation exception to the extra audio consent, names dubbing for entertainment as a permitted use and says personal-voice outputs are watermarked. These pages are Microsoft's statements; the actual Azure terms they refer to ("Service Specific Terms") were not read.
- articleWhat do practitioners say a sound editor must do to prepare tracks before the mix (separation, premixes, scratch mixes, M&E), and what goes wrong when they do not?
Craft-side preparation, as distinct from delivery specs: split dialogue by mic angle; keep families on separate tracks and premix them into manageable elements; have the major effects and clean-ups ready by the first scratch mix; record the full soundscape under dialogue so an M&E exists; decide with the composer, scene by scene, whether music or effects leads. Documented failures include sketches frozen by a director's liking for the scratch mix, a composer/effects logjam in the final mix, and a live-performance compromise (Philadelphia) that left dialogue and aria on one track.
- articleHow do I duck music under dialogue (sidechain or automatic ducking), what parameter values do open manuals state, and what do practitioners do instead of lowering all the music?
Three mechanisms are documented: an offline effect keyed from a control track (Audacity Auto Duck: -12 dB, 0.5 s fades, 1 s maximum pause, -30 dB RMS threshold), a sidechained compressor (ReaComp auxiliary detector, Ardour a-Compressor sidechain; OBS gives 32:1, -36 dB, 100 ms attack, 600 ms release as a starting point for streaming), and an inverted sidechained gate (ReaGate). A feature-film practitioner account adds a third strategy: deliver music as stems and nudge only the competing stem (a loud horn) down a few dB at the dub so the whole score is not dipped. No inspected source validates specific ducking values for film dialogue.
- articleWhat do OBS Studio's libobs docs say about output frame-rate representation and audio/video sync offset, and is a variable-frame-rate recording mode documented?
OBS (libobs at 32.2.2) runs its output at a fixed frame rate expressed as an integer fraction (fps_num/fps_den), set in the UI as a common, integer or fractional value. The UI strings searched have no variable-frame-rate option, so recordings come out constant frame rate, which is convenient for editing.
- articleWhat does C2PA soft binding tell an agent about a watermark or fingerprint, and what does it leave out (robustness, detectability, licence)?
C2PA lets a manifest point to a watermark or fingerprint ('soft binding') so content can be re-matched after the file changes or the manifest is stripped, but a soft binding must never stand in for the cryptographic hard binding. C2PA's registry of soft-binding algorithms lists 28 entries (25 watermarks, 3 fingerprints; 10 cover audio). Entries record only a description, entry date, contact and info URL, with no robustness, detection-rate or licence data. Neither Google's SynthID nor Meta's AudioSeal is listed.
- articleWhat has Netflix's As Recorded Dubbing Script scope of work changed over time, and what does its change log say?
Netflix's As Recorded Dubbing Script scope of work dates its own changes. The script was due five calendar days before the secondary audio until a 17 September 2024 change made it three. A 24 September 2025 change added episode-title and main-title rules (use the approved title; don't translate freely) and removed the separate TTAL required-fields and appendix sections. Edit timing is bounded only by the captures read.
- articleHow did the audio rules in the Post Production Branded Delivery Specification change between page labels 5.0 (2022-12-09), 5.1 (2024-08-03) and 5.2 (2026-06-06)?
Netflix's Branded Delivery Specifications changed Atmos rules in two steps. Version 5.1 (change log 12 September 2023) removed the separate 2.0 stereo M&E. Version 5.2 (change log 21 October 2024) widened the Atmos group labels from four (Dialog, Music, Effects, Narration) to seven (adding Group ADR, Backgrounds, Foley), made 5.1 and 2.0 deliverables optional for Atmos mixes, and replaced the '-2 dBFS true-peak' sentence with a recommendation to set limiters at -2.3. The separate Atmos requirements page (v2.3, 'updated for parity' with the Branded spec) still showed the older wording in an April 2026 capture, and showed the 5.2 wording by October 2026 (curator cross-check). So the conflicts recorded earlier were a lagging page, and current Netflix pages agree.
- articleHow does the ARSC Guide to Audio Preservation separate a preservation transfer from restoration, and what does it say about disc speed and equalisation?
The ARSC Guide to Audio Preservation (CLIR 2015, CC BY-NC-SA 4.0, credited photos excluded) separates preservation from restoration. A preservation transfer reproduces the carrier accurately with aligned machines, the correct stylus and equalization or other decoding as needed. Subjective clean-up is restoration, done only after a preservation copy exists. It defines a preservation master, an access master (which may be restored, e.g. de-noised) and an access copy, and names BWF .wav the de facto archival format. Many '78 rpm' discs should really be played at slightly different speeds (e.g. 76 or 79 rpm), so turntables need incremental speed control.
- articleWhat do EBU R 128 s1, s2, s3, s4, Tech 3343, Tech 3344 and (if available) AES TD1004 add beyond R 128 for production and streaming distribution?
The EBU R 128 family as read in October 2026: R 128 V5 (November 2023) is the base. Supplement s1 (August 2020) sets short-form content (ads, promos) at -23.0 LUFS with short-term loudness no higher than -18.0 LUFS and true peak no higher than -1 dBTP, and says loudness range isn't useful for short-form. Supplement s2 (November 2023) keeps streams at -23.0 LUFS, or an interim distribution level of -20 to -16 LUFS when the broadcaster does the dynamic treatment, citing AES 71 and AES 77. Tech 3343 V4 (November 2023) folds in the streaming, radio and cinematic supplements, and Tech 3344 v2.1 (July 2016) covers distribution and reproduction.