Audio annotation services
Speakers, speech, sounds and background noise labeled in your recordings by native-speaker annotators, with timestamps to the frame your model needs. For speech recognition, voice assistants, call analytics and sound event detection.
Example labels on a ten-second customer call: speakers, overlap, music, sound events and background noise.
- Labels
- Speakers, transcripts, sound events, noise, emotion and language
- Annotators
- Native speakers of the language in the recording
- Formats
- WAV, MP3, FLAC and M4A in; JSON, TextGrid, RTTM or CSV out
- Quality
- Agreement checks and a reviewed sample on every batch
What audio annotation is
Audio annotation is labeling what happens in a recording and when: who is speaking, what they say, where speakers overlap, what other sounds are present and how noisy the background is. Models for speech recognition, call analytics and sound detection learn from these time-stamped labels.
Speech labels need annotators who understand the language, accent and slang in the recording. If you need new recordings made, see speech and audio data.
Audio annotation tasks
Choose the labels your model needs. We'll write the guidelines with you.
Speaker diarization
Who spoke when, with speaker turns and overlapping speech marked.
Transcription for ASR
Verbatim transcripts with timestamps, fillers, false starts and tags for unclear speech. More on transcription.
Sound event detection
Alarms, doors, sirens, laughter, music and other sounds, with start and end times.
Noise and quality
Background noise type and level, clipping, echo and recordings to exclude.
Emotion and intent
Speaker emotion, tone and intent, labeled by people who know the culture.
Language and accent ID
Which language, dialect or accent each segment uses, including code-switching.
How an audio annotation project runs
Timing rules are agreed up front, because they decide how labels line up.
Sample and spec
A sample of your audio, the label set, and timing rules for start and end points.
Pilot
A few hours of audio labeled by several annotators per language.
Agreement check
Label and timing agreement measured, and guidelines updated with you.
Agreement report QAProduction
Batches labeled to your schedule, with a reviewer listening to a sample.
Delivery
Labels in JSON, TextGrid, RTTM, CSV or your tool's format.
Related AI data services
Questions about audio annotation
What is audio annotation?
Audio annotation is labeling recordings with time-stamped information, such as who is speaking, what they say, which sounds occur and how noisy the background is, so the audio can train or test speech and sound models.
Can you label audio in other languages and accents?
Yes. Speech is labeled by native speakers of the language in the recording, and we match annotators to dialects and accents when your project needs it.
Which output formats do you deliver?
JSON, CSV, Praat TextGrid and RTTM are common, or the export format of your annotation tool. Tell us what your pipeline reads.
How do you check timestamp accuracy?
We agree timing rules in the guidelines, such as where a segment starts and ends, then measure timing agreement between annotators in the pilot and check a sample of every batch.
Send us sample audio
Send a few recordings and the labels you need, or book a call. We label a pilot and report timing and label agreement.
- Phone
- +44 20 3290 0116
- Phone hours
- Weekdays 9:30 to 16:00, weekends 9:30 to 12:00, UK time