I want to build an app for Indian-language users. What do speech and language APIs cost?

The Indian specialists publish rupee prices. Sarvam charges Rs 0.49 a minute for speech-to-text across 22 Indian languages, Rs 1 per 1,000 characters for its Bulbul V4 Flash voices, Rs 2 per 1,000 characters for translation, and Rs 15 per million input tokens for its 105B model, with 100 free credits to test; a phone voice agent starts at Rs 3 a minute at high volume. To host a model yourself, Sarvam-105B is Apache 2.0 and BharatGen's Param2 is open too. Bhashini offers its APIs to organisations that register. Global models work, and ElevenLabs voices cover Hindi, Tamil and more, but Sarvam reports other open models' tokenizers need more tokens per word of Indian text, so price your real messages, in script and in Hinglish, before you commit.