Reference

Audio & Transcription Glossary

Every term you will meet working with audio and transcripts — 124 plain-English definitions across speech recognition, captions, audio fundamentals, engineering, podcasting and voice AI. Each entry leads with a quotable definition and links the concepts around it.

Transcription & speech recognition (31)

Acoustic modelAn acoustic model is the component of a speech-recognition system that maps audio features to the so… Automatic speech recognition (ASR)Automatic speech recognition (ASR) is the technology that converts spoken language in audio into mac… Batch transcriptionBatch transcription processes complete, already-recorded audio files, letting the system use full co… Beam searchBeam search is a decoding strategy where a speech-recognition model keeps several candidate transcri… Character error rate (CER)Character error rate (CER) measures speech-recognition accuracy at the character level rather than t… Clean verbatimClean verbatim is a transcription style that preserves every meaningful word while removing filler s… Code-switchingCode-switching is when a speaker alternates between two or more languages within a conversation or a… Confidence scoreA confidence score is the speech recognizer's own estimate of how likely each word or segment is cor… DictationDictation is speech recognition used as a writing tool: you speak, punctuation and formatting are in… End-to-end speech recognitionEnd-to-end speech recognition uses a single neural network that maps audio directly to text, replaci… Forced alignmentForced alignment is the process of taking an existing transcript and an audio recording and computin… Greedy decodingGreedy decoding is the fastest way for a speech model to produce text: at every step it emits the si… Hallucination (in transcription)In speech recognition, hallucination is when the model outputs fluent text that was never spoken — t… Intelligent verbatimIntelligent verbatim is the most edited transcription style: fillers, repetitions and irrelevant asi… Inverse text normalization (ITN)Inverse text normalization converts spoken forms into written conventions — 'twenty twenty six' beco… Language model (in ASR)In speech recognition, a language model supplies knowledge of which word sequences are likely, letti… Multilingual speech recognitionMultilingual speech recognition uses one model trained across many languages — Whisper's published l… Punctuation restorationPunctuation restoration is the step that turns raw recognized words into readable text by inserting … Quantization (of speech models)Quantization shrinks a neural speech model by storing its weights at lower numeric precision — for e… Real-time factor (RTF)Real-time factor (RTF) measures transcription speed as processing time divided by audio duration: an… Speaker diarizationSpeaker diarization is the process of determining who spoke when in a recording — segmenting audio b… Speaker identificationSpeaker identification determines which known person is speaking by matching voices against enrolled… Speaker labelsSpeaker labels are the per-turn tags in a transcript — 'Speaker 1', 'Interviewer', or a real name — … Streaming speech recognitionStreaming speech recognition transcribes audio as it arrives, emitting words within moments of them … TimestampingTimestamping marks transcript text with the time it occurs in the recording — per segment, per speak… Transcript editingTranscript editing is the human pass after automatic transcription — fixing misheard words, correcti… TranscriptionTranscription is the process of converting spoken audio into written text, either by a human typist … Verbatim transcriptionVerbatim transcription captures speech exactly as uttered — including false starts, repetitions, fil… Voice activity detection (VAD)Voice activity detection (VAD) is the technique of finding which parts of an audio signal contain sp… Word error rate (WER)Word error rate (WER) is the standard accuracy metric for speech recognition: the number of word sub… Word-level timestampsWord-level timestamps attach a start and end time to every individual word in a transcript, rather t…

Captions & subtitles (18)

Burned-in captionsBurned-in (or hardcoded) captions are rendered into the video frames themselves during export, becom… Caption line lengthCaption line length is the maximum characters per caption line — commonly 32 in broadcast tradition,… Caption positioningCaption positioning is where captions sit on screen — bottom-center by default, moved when they woul… Caption reading speedCaption reading speed measures how fast text asks viewers to read, commonly in characters per second… CEA-608 captionsCEA-608 is the legacy North American broadcast standard for closed captions — the 'line 21' captions… CEA-708 captionsCEA-708 is the digital-television caption standard in North America, supporting multiple windows, fo… Characters per second (CPS)Characters per second (CPS) is the standard measure of caption reading speed: total characters in a … Closed captionsClosed captions are timed text that can be turned on or off by the viewer, written for deaf and hard… Forced narrative subtitlesForced narrative (forced subs) are subtitles that appear even when subtitles are 'off' — translating… Karaoke-style captionsKaraoke-style captions highlight each word at the moment it is spoken, the style popularized by soci… Live captioningLive captioning produces captions in real time for broadcasts, streams, meetings and events, via str… Open captionsOpen captions are permanently burned into the video image — every viewer sees them, no player suppor… RespeakingRespeaking is a live-captioning method where a trained captioner listens to the broadcast and re-dic… SDH (Subtitles for the Deaf and Hard-of-hearing)SDH subtitles combine the translation role of subtitles with the accessibility content of captions —… SRT fileSRT (SubRip Text) is the simplest and most widely supported subtitle format: numbered blocks of star… Subtitle timingSubtitle timing (spotting) is deciding when each caption appears and disappears — synchronized to sp… SubtitlesSubtitles are timed text displayed over video showing what is being said, primarily for viewers who … WebVTT fileWebVTT (VTT) is the web-native subtitle format used by HTML5 video, extending SRT-style cues with st…

Audio fundamentals (25)

AliasingAliasing is the distortion produced when frequencies above the Nyquist limit are sampled anyway, fol… AmplitudeAmplitude is the magnitude of an audio signal's oscillation — the physical correlate of loudness, di… Bit depthBit depth is how many bits describe each audio sample — 16-bit (CD) gives about 96 dB of dynamic ran… BitrateBitrate is the amount of data used per second of audio, in kilobits per second (kbps) — the main qua… Channels (mono, stereo, multichannel)Channels are the independent audio streams in a recording: mono has one, stereo two, surround format… ClippingClipping is the distortion that occurs when a signal exceeds the maximum level a system can represen… CodecA codec (coder-decoder) is the algorithm that encodes audio for storage or transmission and decodes … Container formatA container format is the file wrapper that packages encoded audio (and video, captions, metadata) —… dBFS (decibels relative to full scale)dBFS measures digital audio level against the maximum a file can represent: 0 dBFS is the ceiling, a… Decibel (dB)The decibel is the logarithmic unit of audio level: +6 dB is roughly double the signal amplitude, an… DitheringDithering adds a precisely shaped trace of noise when reducing bit depth, converting quantization di… Dynamic rangeDynamic range is the span between the quietest and loudest parts of audio — of a performance, a reco… FrequencyFrequency is how many times per second a sound wave cycles, measured in hertz (Hz) — perceived as pi… Frequency responseFrequency response describes how a microphone, speaker or system reproduces different frequencies — … Gain stagingGain staging is managing signal level at every point in an audio chain — mic preamp, interface, soft… HeadroomHeadroom is the safety margin between your audio's peaks and the maximum level the system can handle… Lossless compressionLossless compression (FLAC, ALAC) shrinks audio while preserving it bit-for-bit — decoding restores … Lossy compressionLossy compression shrinks audio by permanently discarding information judged least audible — the app… Noise floorThe noise floor is the constant background level present in a recording when nobody is speaking — ro… Nyquist theoremThe Nyquist–Shannon theorem states that a digital system can perfectly capture frequencies up to hal… PCM (pulse-code modulation)PCM is the standard representation of uncompressed digital audio: the signal is sampled at a fixed r… ResamplingResampling converts audio from one sample rate to another — 48 kHz video audio to 16 kHz for a speec… Sample rateSample rate is how many times per second an audio signal is measured when converting sound to digita… Signal-to-noise ratio (SNR)Signal-to-noise ratio compares desired audio to background noise, in decibels — higher SNR means cle… WaveformA waveform is the visual plot of an audio signal's amplitude over time — the shape you scrub through…

Audio engineering (21)

Compression (dynamics)A compressor automatically reduces the level of loud moments, narrowing the gap between quiet and lo… CrossfadeA crossfade overlaps the end of one audio region with the start of the next, fading one out while th… De-esserA de-esser is a frequency-targeted compressor that tames sibilance — the piercing 's' and 'sh' energ… Delay (effect)Delay repeats a signal after a set time — the building block of echo effects and, at short times, of… Equalization (EQ)Equalization adjusts the balance of frequencies in audio — cutting rumble, taming harshness, adding … Fade in / fade outA fade gradually raises audio from silence (fade-in) or lowers it to silence (fade-out), replacing a… High-pass filterA high-pass filter removes frequencies below a chosen cutoff, letting highs pass — the near-universa… LimiterA limiter is a compressor with an effectively absolute ceiling, guaranteeing audio never exceeds a s… Loudness normalizationLoudness normalization adjusts audio to a target perceived loudness, measured in LUFS, rather than t… LUFSLUFS (Loudness Units relative to Full Scale) measures perceived loudness — integrating level over ti… MasteringMastering is the final processing pass on a finished mix — overall tone, loudness to target, true-pe… MixingMixing combines multiple recorded tracks into one balanced whole — setting levels, EQ, compression a… Noise gateA noise gate mutes audio when it falls below a threshold, silencing the background between phrases w… Normalization (peak)Peak normalization scales an entire recording so its loudest sample hits a target level — a uniform … PanningPanning places a sound in the stereo field between left and right speakers, creating spatial separat… PlosivesPlosives are the bursts of air from 'p' and 'b' sounds that hit a close microphone as a low-frequenc… Proximity effectProximity effect is the bass boost directional microphones add as the source gets close — the 'radio… ReverbReverb is the persistence of sound as it reflects around a space — added intentionally for music, fo… Room toneRoom tone is the 'silence' of a recording space — its constant ambient character — deliberately reco… SibilanceSibilance is the high-frequency energy of 's', 'sh' and 'ch' sounds, concentrated around 5–8 kHz, wh… True peakTrue peak measures the actual analog waveform maximum after digital-to-analog conversion, which can …

Podcast & production (14)

AudiogramAn audiogram is a short video made from audio — waveform animation, captions and artwork over a clip… Back catalogA back catalog is a show's archive of past episodes — often hundreds of hours whose value compounds … Double-ender (remote recording)A double-ender records each remote participant locally on their own machine, then syncs the tracks i… Dynamic ad insertion (DAI)Dynamic ad insertion assembles the ads in a podcast episode at download time, so different listeners… ID3 tagsID3 tags are the metadata block inside MP3 files — title, artist, artwork, chapters — read by player… Multitrack recordingMultitrack recording keeps each source — each voice, each instrument — as its own synchronized track… Podcast chaptersPodcast chapters divide an episode into named, timestamped sections listeners can skip between, deli… Podcast hostingPodcast hosting stores a show's audio and serves its RSS feed, handling the bandwidth of thousands o… Podcast loudness targetThe podcast loudness convention is approximately −16 LUFS integrated for stereo (−19 for mono) with … Podcast transcriptA podcast transcript is the full text of an episode, published for accessibility, search and reuse —… Pre-roll / mid-roll / post-rollPre-roll, mid-roll and post-roll name ad positions in an episode: before content, inside it, and aft… Remote recordingRemote recording captures conversations whose participants are in different places — from recording … RSS feed (podcasting)A podcast RSS feed is the XML file that defines a show — its metadata and an entry per episode with … Show notesShow notes are the written companion to a podcast episode — summary, links, timestamps, guest info —…

Voice & AI audio (15)

Audio classificationAudio classification labels what a sound is — speech, music, applause, glass breaking — without tran… DereverberationDereverberation removes room echo from recordings, estimating and subtracting the reflected sound to… Fundamental frequency (F0)Fundamental frequency (F0) is the rate of vocal-fold vibration — perceived as the pitch of a voice, … Keyword spottingKeyword spotting detects specific words or phrases in audio without transcribing everything — the li… Mel spectrogramA mel spectrogram is the time-frequency picture of audio that speech models actually consume — a spe… MFCC (Mel-frequency cepstral coefficients)MFCCs are a compact numeric summary of a sound's spectral shape on the mel scale — the dominant audi… PhonemeA phoneme is the smallest sound unit that distinguishes words in a language — English has roughly 44… ProsodyProsody is the melody of speech — pitch movement, rhythm, stress and pauses — carrying meaning beyon… Source separationSource separation splits mixed audio into components — vocals from music, one speaker from another —… Speaker embeddingA speaker embedding is a compact numeric vector representing the characteristics of a voice, such th… Speech enhancementSpeech enhancement cleans a voice recording — suppressing noise, reducing reverb, restoring clarity … Text-to-speech (TTS)Text-to-speech synthesizes spoken audio from written text — the inverse of speech recognition — with… Voice cloningVoice cloning creates a synthetic voice that imitates a specific real person, sometimes from minutes… Voice conversionVoice conversion transforms one person's recorded speech to sound like another voice while preservin… Wake word detectionWake word detection is the tiny, always-on model that listens for a trigger phrase ('Hey Siri') and …