[Feature]: pls add support to this ASR models

[Excurl]

Is your feature request related to a problem?

please add support to the indic models, unfortunately none of the dictation app supports it... :)

https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual

am specifically looking for malayalam, which is at https://huggingface.co/ai4bharat/indicconformer_stt_ml_hybrid_ctc_rnnt_large , but its in .nemo architecture, if it possible to add after onnxing or add already onnxed multilingual above ( https://huggingface.co/ai4bharat/indic-conformer-600m-multilingual )

Describe the solution you'd like

malayalam language support

Describe alternatives you've considered

No response

Feature Scope

None

How important is this feature to you?

None

Additional context

No response

Contribution

• I'd like to work on this feature myself

[watzon]

Thanks for this! On the ROADMAP!

[Excurl]

any scope on this?

[watzon]

Thanks for the detailed feature request! After investigating the models you mentioned:

• ai4bharat/indic-conformer-600m-multilingual - ONNX format with custom Python code requirements, no native Swift/macOS runtime available
• ai4bharat/indicconformer_stt_ml_hybrid_ctc_rnnt_large - NVIDIA NeMo format, not practical for macOS integration

Good news: OpenAI's Whisper models (which Pindrop already uses) were trained on significant multilingual data that includes Indic languages. Whisper large-v3 already has support for Malayalam, Hindi, and all 22 Indian languages recognized in your request.

Planned for next release: Instead of adding a custom backend, I will add Malayalam and Hindi as supported languages in Pindrop's UI, so you can:

1. Have the app interface in your language
2. Dictate in Malayalam/Hindi with proper language selection

This way you get native Indic language support without the complexity of integrating a completely different model architecture.

If you want better accuracy for Malayalam specifically, fine-tuned Whisper models like vrclc/Whisper-medium-Malayalam exist and could be added in a future release once CoreML conversion tooling is more accessible.