What I'm asking for
Following up on a Discord chat with YC (PM), who suggested I open this. Thanks for adding Tamil under Sri Lanka. Could you add Indian (Tamil Nadu) Tamil as a second selectable variant, locale ta-IN, alongside the Sri Lankan one you just shipped?
Why a second variant
I speak Tamil. Sri Lankan Tamil and Tamil Nadu (Indian) Tamil are not interchangeable in speech. The written form is close, but the spoken pronunciation, prosody and everyday register differ enough that an agent tuned to the Sri Lankan variant sounds noticeably off to a Tamil Nadu listener. The reverse is true too. Providers model them as separate locales for that reason (ta-LK vs ta-IN). Tamil Nadu is the larger audience, so the Indian variant covers most Tamil speakers a call would reach.
Suggested ASR/TTS provider
You asked for a suggested provider, so here are the ones that expose the Indian Tamil locale, checked against their own docs today.
Sarvam AI is the closest fit for this variant. It is built for Indian languages and accents rather than adapted from English. Its Tamil is ta-IN. It is tuned for telephony (8kHz, sub-250ms streaming, marketed for voice agents and IVR), which matches a phone-call product.
If you would rather stay on a mainstream engine, both of these expose ta-IN as well:
One thing worth noting: across Google and Azure, ta-IN is the only Tamil locale exposed for recognition (neither lists ta-LK), so the Indian variant is the more broadly supported one. Enabling it should be low-risk on whatever engine you are already using.
Happy to test
If you enable ta-IN, I can run an end-to-end call in Tamil Nadu Tamil and report back how the recognition and the voice sound, the same way the Finland route was validated in #74.
What I'm asking for
Following up on a Discord chat with YC (PM), who suggested I open this. Thanks for adding Tamil under Sri Lanka. Could you add Indian (Tamil Nadu) Tamil as a second selectable variant, locale
ta-IN, alongside the Sri Lankan one you just shipped?Why a second variant
I speak Tamil. Sri Lankan Tamil and Tamil Nadu (Indian) Tamil are not interchangeable in speech. The written form is close, but the spoken pronunciation, prosody and everyday register differ enough that an agent tuned to the Sri Lankan variant sounds noticeably off to a Tamil Nadu listener. The reverse is true too. Providers model them as separate locales for that reason (
ta-LKvsta-IN). Tamil Nadu is the larger audience, so the Indian variant covers most Tamil speakers a call would reach.Suggested ASR/TTS provider
You asked for a suggested provider, so here are the ones that expose the Indian Tamil locale, checked against their own docs today.
Sarvam AI is the closest fit for this variant. It is built for Indian languages and accents rather than adapted from English. Its Tamil is
ta-IN. It is tuned for telephony (8kHz, sub-250ms streaming, marketed for voice agents and IVR), which matches a phone-call product.If you would rather stay on a mainstream engine, both of these expose
ta-INas well:ta-IN: https://docs.cloud.google.com/speech-to-text/docs/languagesta-INfor speech to text: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-supportOne thing worth noting: across Google and Azure,
ta-INis the only Tamil locale exposed for recognition (neither liststa-LK), so the Indian variant is the more broadly supported one. Enabling it should be low-risk on whatever engine you are already using.Happy to test
If you enable
ta-IN, I can run an end-to-end call in Tamil Nadu Tamil and report back how the recognition and the voice sound, the same way the Finland route was validated in #74.