Bodhi API Platform
Speech infrastructure for Indian languages
Transcribe audio and synthesize speech across eleven models and ten languages — as a one-shot request or a live bidirectional stream. One API key, credits metered per use, no minimum plan.
curl -X POST https://api.bodhi.example.com/api/asr/batch \
-H "Authorization: Bearer $BODHI_API_KEY" \
-F "model=hi-banking-v2-8khz" \
-F "audio_file=@recording.wav"Capabilities
Speech to Text
Transcribe a WAV file in one call, or stream audio over a WebSocket and read transcripts as the words are still being spoken. Segment timings, confidence scores, and word-level timestamps come back with every result.
11 models · 8 languages · batch + streaming
Text to Speech
Synthesize a block of text in one call, or push text into a WebSocket and receive audio chunks as they are generated — start playing the first sentence before the last one exists.
57 voices · 10 languages · batch + streaming
Why build on it
Indian languages, telephony-first
Models trained per language and per domain at 8 kHz — the rate real phone calls arrive at. Every synthesis voice is available in all ten languages under one id.
Streaming both ways
Real WebSocket endpoints for transcription and synthesis, each metered live. Compose them into a full speech-in, speech-out loop on one key.
Metering that fails safe
Batch requests are priced and checked before the engine is contacted, so an unaffordable call is rejected in milliseconds rather than billed after the fact.
Per-key budgets
Rate and concurrency limits are enforced per key, and transcription and synthesis hold separate budgets — saturating one never starves the other.