Audio annotation and transcription services for speech AI
Send us recordings you already have. Native-speaker annotators transcribe and label them for ASR, TTS and voice AI under a written guideline, quality is measured against a threshold you agree, and delivery follows your schema.
Annotation layers and output formats
Speech annotation records what was said, when and by whom. Audio labeling adds the rest: non-speech events, intent, sentiment and emotion. Scope only the layers your model will use.
| Layer | What it covers | Typical formats |
|---|---|---|
| Verbatim transcription | Human or ASR-assisted transcripts with agreed rules for casing, numbers, disfluencies and domain terms | JSON, JSONL, CSV |
| Timestamps and segmentation | Word- or segment-level timestamps with agreed boundary rules and tolerances | JSON, SRT, VTT |
| Speaker turns and diarization | Anonymous speaker labels with rules for overlap, turns and unknown speakers | JSON, JSONL, CSV |
| Non-speech events | Background noise, music, laughter, silence and channel conditions, labeled in place | JSON, JSONL |
| Intent, sentiment and emotion | Utterance-level labels applied against a defined tag set | JSON, CSV |
| Custom schema | Your own labels, metadata fields and edge cases, mapped from your guideline | Custom manifest |
Formats are options, not defaults. The written scope confirms the format for each layer.
Guideline first, then measured quality
Native-speaker annotators
Annotators are native speakers of the language they label, across the 60 languages and dialects in our catalog and more on request. For other languages, we confirm reviewer feasibility before quoting.
Written guideline and pilot
We agree a written guideline before production: casing, numbers, disfluencies, overlap, unknown speakers and domain terms. A pilot batch can calibrate annotators and close gaps in the guideline before full production.
Measured QA
We measure inter-annotator agreement on duplicated items, and word error rate against a gold reference where transcription accuracy matters. The scope states the sample, the threshold and what triggers rework.
Pricing and delivery
Priced per audio hour, item or batch, quoted after we review a sample of your audio. Language, audio difficulty, label layers and QA depth set the rate. Delivery follows your schema.
Frequently asked questions
Can you annotate audio we already have?
Yes. We transcribe and label buyer-provided recordings as a standalone service. Send a representative sample, and the written scope confirms the labels, languages, QA method, output format and price before production starts.
Is voice annotation the same as transcription?
Transcription is one layer of it. Voice annotation can also mark when each word or segment is spoken, who is speaking, non-speech sounds such as music or laughter, and the intent, sentiment or emotion of each utterance. Label layers set the rate, so the written scope lists only the layers your model needs.
Which annotation layers can you deliver?
Verbatim transcription, word- or segment-level timestamps, speaker turns and diarization labels, non-speech events such as noise, music and laughter, utterance-level intent, sentiment and emotion labels, and custom fields mapped from your own guideline. Common formats are JSON, JSONL, CSV, SRT and VTT, or a custom manifest.
How do you measure annotation quality?
Against numbers agreed before production. Inter-annotator agreement is measured on duplicated items, and transcription accuracy as word error rate against a gold reference where relevant. The written scope states the sample, the threshold and what triggers rework.
How is audio annotation priced?
Per audio hour, item or batch. The rate depends on language, audio quality, label layers and QA depth, so we quote after reviewing a sample of your audio.
How is our audio handled?
The contract sets who can access your audio, where it is stored and how long it is kept. Storage and access requirements are agreed before work begins.
Tell us about your audio
Share the languages, hours, audio type and labels you need. We reply within one business day and quote after reviewing a sample.
Need the audio recorded first? We can run custom collection and annotation as one project.