Transcription · ASR Model Training
Transcription Services for ASR Model Training
Verbatim and clean-read transcription of Indian-language audio by native speakers of the target variety, delivered against a written style guide with measured agreement. Applied to asr model training, the specification is driven by one thing: word error rate overall and per dialect.
- Service
- Transcription
- 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 |
|---|---|
| Modes | Verbatim, clean verbatim, or clean-read |
| Timestamps | Utterance or word level |
| Script | Native script, romanised, or dual-script delivery |
| Tags | Speaker, noise, overlap, unintelligible, code-switch |
| Accuracy target | 99% on clean studio audio; agreed per-project on field audio |
Process
- Style guide authoring and client sign-off
- Transcriber calibration and agreement measurement
- First pass transcription
- Independent second-pass review
- Automated rule checks and delivery
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
The style guide is the deliverable that determines whether your data is usable. It is written first and shipped with the data.
Deliverables
- Transcripts in your schema
- Style guide
- QA and agreement report
Frequently asked
Is transcription 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 transcription for asr model training
Send the metric you need to move.