ASR
Diverse, hard-condition speech with verbatim and clean transcripts to train and test recognition.
Voice & speech AI data
Speech data that sounds like real people.
Scripted and natural recordings from diverse speakers, with careful transcription and the labels voice agents need to understand intent — not just words.
What it is
Recordings, transcripts, and the meaning behind them.
Voice models fail on the speech they rarely hear: accents, noisy rooms, far-field mics, real conversations. We record it on purpose, to your spec.
- Scripted, conversational, and role-play recordings
- Speakers matched to your demographic spec
- Accents, multilingual, call-center, far-field, and emotional-range speech
- Speaker and acoustic-condition metadata on every file
- Voice-use rights documented for every speaker
Studio and natural speechVoice
Who it’s for
Built for teams shipping speech and voice agents.
TTS
Scripted and expressive recordings with speaker metadata for synthesis work.
Voice agents
Intent, sentiment, and tone labels plus voice-agent evaluations and human preference ratings.
How it works
Consent, transcription, and QA done properly.
Record
Contributors record scripted or natural speech following written guidelines and device specs agreed in scoping.
Consent & privacy
Every speaker signs a rights agreement with documented voice-use rights, and PII is redacted before data moves downstream.
Transcribe, label & QA
Trained transcribers and labelers produce timestamped transcripts and intent, sentiment, and tone labels, then multi-pass QA reviews every file.
What’s in a record
Every file ships with the labels a model needs.
- Verbatim and clean transcripts with timestamps
- Intent, sentiment, and tone labels
- Speaker metadata matched to your spec
- Acoustic-condition metadata
- Consent records and QA status
- Optional voice-agent eval scores and preference ratings
AudioWAV / FLAC
LabelsJSON · CSV
TasksASR, TTS, agent evals
DeliveryS3, Hugging Face, or your bucket
Every delivered recordYour schema
- Source
- Domain, task, modality, expert role & experience
- Content
- Media or text, with timestamps where relevant
- Labels
- Steps, actions, objects, intents, rubrics, scores
- Expert notes
- Rationale, technique, edge cases
- Privacy
- Face blur & PII redaction flags
- Consent
- Contributor agreement & site/owner consent
- QA
- Review passes and approval status
- Format
- JSON · JSONL · CSV · parquet · COCO
FAQ
Speech & voice data: common questions
What kinds of speech data do you collect?
Scripted, conversational, and role-play recordings from diverse speakers, including accents, multilingual, call-center, far-field, and emotional-range speech.
What labels come with the audio?
Verbatim and clean transcripts with timestamps, plus intent, sentiment, and tone labels and speaker and acoustic-condition metadata.
Can you evaluate voice agents?
Yes. We run voice-agent evaluations and collect human preference ratings for ASR, TTS, and agent tasks.
How are speakers chosen?
Speakers are matched to your demographic spec, agreed during scoping.
How is consent handled?
Every contributor signs a rights agreement, and voice-use rights are documented for each speaker.
What formats and delivery do you support?
WAV or FLAC audio with JSON or CSV labels, delivered to S3, a private Hugging Face repo, or your bucket, with a data card and schema docs.
Start with a sample
See the data before you commit.
Tell us what your model needs to learn. We’ll scope a pilot and send a sample set in your schema.