vakyansh-wav2vec2-tamil-tam-250
Harveenchadha/vakyansh-wav2vec2-tamil-tam-250
A popular open speech-to-text model, with 2.2M downloads a month. gigarouter benchmarks and hosts it as an OpenAI-compatible API.
about this model
Harveenchadha/vakyansh-wav2vec2-tamil-tam-250 is an automatic speech recognition (ASR) model for Tamil, fine-tuned from the multilingual CLSRIL-23 pretrained checkpoint. It is trained on 4200 hours of labelled speech data and requires 16 kHz input audio.
Key Strengths
- Fine-tuned from a multilingual foundation model, enabling robust Tamil speech recognition.
- Delivered without an external language model; Word Error Rate can be further reduced by integrating a language model if needed.
- Suitable for direct inference or as a component in a larger pipeline.
Benchmark Performance
| Dataset | Metric | Score | Notes |
|---|---|---|---|
| Common Voice (Tamil test set) | Word Error Rate (WER) | 53.64% | Without a language model |
The reported WER is from the model’s standalone evaluation; performance may improve when combined with a language model.
Intended Use
This model is designed for developers deploying Tamil ASR in applications where a dedicated, fine-tuned wav2vec 2.0 model is preferred. It works best as a hosted API via gigarouter, eliminating the need for manual model loading and infrastructure management.
We're benchmarking and onboarding vakyansh-wav2vec2-tamil-tam-250 as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.