Call-centre speech · ASR Model Training
Call Centre Speech Data for ASR Model Training
Simulated and consented real-world contact-centre audio in Indian languages, recorded over telephony-grade channels so it matches the bandwidth your production system actually sees. Applied to asr model training, the specification is driven by one thing: word error rate overall and per dialect.
- Service
- Call-centre speech
- Use case
- ASR Model Training
- Primary metric
- Word error rate overall and per dialect
Required data profile
- Hundreds to thousands of hours of verbatim-transcribed speech
- Wide speaker diversity: age, gender, region, education, recording condition
- Speaker-disjoint train/dev/test splits
Technical specification
| Parameter | Standard |
|---|---|
| Channel | 8 kHz narrowband telephony plus 48 kHz reference where required |
| Codec | G.711 / Opus simulated to match your stack |
| Structure | Agent and customer on separate channels |
| Scenarios | Billing, delivery, KYC, recharge, collections, support, sales |
| Emotion | Neutral, frustrated and escalated variants on request |
Process
- Scenario library built from your call taxonomy
- Agent-side speakers briefed on your script and tone
- Recording over the specified codec path
- Transcription with intent and outcome labels
- Delivery with per-scenario counts
Metrics this feeds
- Word error rate overall and per dialect
- Entity error rate on names and numbers
- Code-switch token accuracy
Failure modes to design out
- Read-speech-only corpora that do not transfer to spontaneous audio
- Speaker leakage across splits inflating reported accuracy
- Normalised-only transcripts with the raw text discarded
Narrowband is simulated at capture time, not by downsampling clean studio audio, which would leave unrealistic artefacts.
Deliverables
- Dual-channel call audio
- Verbatim transcripts
- Intent, outcome and sentiment labels
- Scenario coverage matrix
Frequently asked
Is call-centre speech the right service for asr model training?
It covers hundreds to thousands of hours of verbatim-transcribed speech. Most asr model training programmes combine it with at least one other service; we will say so in the scope rather than selling one line item.
What languages are available?
All 14 languages in the network plus Indian English accent bands.
How is the evaluation set handled?
Collected first, from speakers disjoint from the training cohort, so improvement is measurable.
Scope call-centre speech for asr model training
Send the metric you need to move.