Models and Voice Reference
This reference catalogs the available models, service IDs, voice identifiers, and provider parameters across all supported cloud and local engines in indic-language-utils.
Sarvam AI
Sarvam AI provides high-performance generative models for Indian languages via API. Set SARVAM_API_KEY to enable.
Translation Models
| Model ID | Capability | Description | Supported Languages |
|---|---|---|---|
sarvam-translate:v1 |
Translation | Standard multi-way Indic translation model (Default) | en, hi, ta, te, bn, mr, gu, kn, ml, or, pa |
mayura:v1 |
Translation | High-accuracy Indic translation model | en, hi, ta, te, bn, mr, gu, kn, ml, or, pa |
Environment variable override: SARVAM_MODEL
Speech to Text Models
| Model ID | Capability | Description | Limits / Notes |
|---|---|---|---|
saarika:v1 |
Speech to Text | Standard Indic ASR model (Default) | Synchronous REST, max 30s clips |
saarika:v2 |
Speech to Text | Enhanced vocabulary and punctuation | Synchronous REST, max 30s clips |
saarika:flash |
Speech to Text | Ultra low-latency transcription | Synchronous REST, fast response |
Environment variable override: SARVAM_STT_MODEL_ID
Text to Speech Models and Voices
| Model ID | Generation / Status | Supported Parameters | Notes |
|---|---|---|---|
bulbul:v3 |
Flagship 2026 (Default) | speaker, pace, temperature |
Ultra-natural prosody, 35+ speakers, 11 languages |
bulbul:v4-flash |
Low-Latency / Conversational | speaker, pace |
Sub-150ms TTFT, intent-specialized conversational voices |
bulbul:v3-beta |
Experimental Preview | speaker, pace, temperature |
Early release of Bulbul v3 improvements |
bulbul:v2 |
Stable Legacy | speaker, pace, pitch, loudness |
Previous generation TTS model |
Available Speaker Voices for bulbul:v3:
aditya,ritu,ashutosh,priya,neha,rahul,pooja,rohan,simran,kavya,amit,dev,ishita,shreya,ratan,varun,manan,sumit,roopa,kabir,aayan,shubh,advait,anand,tanya,tarun,sunny,mani,gokul,vijay,shruti,suhani,mohit,kavitha,rehan,soham,rupali
Legacy Voices for bulbul:v2:
shubh,arvind,amartya,priya,meera,pavithra
Domain-Specialized Voices for bulbul:v4-flash:
- Conversational / Assistant:
aayan_hi_conversational,amit_hi_conversational,kavya_hi_conversational,simran_hi_conversation,sanchita_hi_assistant,dev_en_conversational,simran_en_conversation - Hinglish Bilingual:
ishita_enhi_companion,shubh_enhi_companion,shubh_enhi_banking,simran_enhi_customer - Customer Support / Banking:
shubh_hi_customer,sanchita_hi_banking,ishita_hi_banking,aditya_hi_sales,jaspal_pa_banking - Regional Narrators:
bappa_bn_conversation(Bengali),pooja_gu_conversational(Gujarati),chaitra_kn_conversation(Kannada),mrunal_mr_narration(Marathi),anand_pa_conversation(Punjabi),gokul_ta_narration(Tamil),kavitha_te_conversation(Telugu)
Environment variable override: SARVAM_TTS_MODEL_ID
Bhashini National Language Translation Mission
Bhashini (backed by AI4Bharat, Bodhan.AI, and the Government of India) uses pipeline serviceId values to route tasks. Set BHASHINI_API_KEY to enable.
Translation Service IDs
| Service ID | Capability | Description |
|---|---|---|
ai4bharat/indictrans-v2-all-gpu--t4 |
Translation | IndicTrans2 universal model supporting all 22 Eighth Schedule languages and English (Default) |
ai4bharat/indictrans-v2-indo_aryan-gpu--t4 |
Translation | IndicTrans2 specialized Indo-Aryan language branch |
ai4bharat/indictrans-v2-dravidian-gpu--t4 |
Translation | IndicTrans2 specialized Dravidian language branch |
Environment variable override: BHASHINI_TRANSLATION_SERVICE_ID
Speech to Text Service IDs
| Service ID | Provider / Architecture | Language | Description |
|---|---|---|---|
ai4bharat/conformer-hi-gpu--t4 |
AI4Bharat Conformer | Hindi (hi) |
AI4Bharat Hindi Conformer ASR (Default for Hindi) |
bhashini/bodhan/asr-transcribe-flex |
Bodhan.AI | Multilingual | Flexible ASR model covering 22+ languages and regional Indian dialects |
bhashini/bodhan/asr-transcribe-core |
Bodhan.AI | Multilingual | Core high-accuracy multilingual ASR transcription pipeline |
ai4bharat/conformer-ta-gpu--t4 |
AI4Bharat Conformer | Tamil (ta) |
AI4Bharat Tamil Conformer ASR |
ai4bharat/conformer-te-gpu--t4 |
AI4Bharat Conformer | Telugu (te) |
AI4Bharat Telugu Conformer ASR |
ai4bharat/conformer-bn-gpu--t4 |
AI4Bharat Conformer | Bengali (bn) |
AI4Bharat Bengali Conformer ASR |
ai4bharat/conformer-mr-gpu--t4 |
AI4Bharat Conformer | Marathi (mr) |
AI4Bharat Marathi Conformer ASR |
ai4bharat/conformer-gu-gpu--t4 |
AI4Bharat Conformer | Gujarati (gu) |
AI4Bharat Gujarati Conformer ASR |
ai4bharat/conformer-kn-gpu--t4 |
AI4Bharat Conformer | Kannada (kn) |
AI4Bharat Kannada Conformer ASR |
ai4bharat/conformer-ml-gpu--t4 |
AI4Bharat Conformer | Malayalam (ml) |
AI4Bharat Malayalam Conformer ASR |
ai4bharat/conformer-pa-gpu--t4 |
AI4Bharat Conformer | Punjabi (pa) |
AI4Bharat Punjabi Conformer ASR |
ai4bharat/conformer-or-gpu--t4 |
AI4Bharat Conformer | Odia (or) |
AI4Bharat Odia Conformer ASR |
ai4bharat/whisper-medium-en--gpu--t4 |
AI4Bharat IndicWhisper | English (en) |
AI4Bharat IndicWhisper English medium model |
ai4bharat/whisper-all-gpu--t4 |
AI4Bharat IndicWhisper | Multilingual | AI4Bharat IndicWhisper multilingual model |
Environment variable override: BHASHINI_STT_MODEL_ID
Text to Speech Service IDs
| Service ID | Provider / Architecture | Description | Supported Parameters |
|---|---|---|---|
ai4bharat/indic-tts-coqui-indo_aryan-gpu--t4 |
AI4Bharat Indic-TTS | Indic-TTS for Indo-Aryan languages (Default) | gender, voiceId, samplingRate |
ai4bharat/indic-tts-coqui-dravidian-gpu--t4 |
AI4Bharat Indic-TTS | Indic-TTS for Dravidian languages | gender, voiceId, samplingRate |
ai4bharat/indic-tts-coqui-misc-gpu--t4 |
AI4Bharat Indic-TTS | Indic-TTS for other scheduled languages | gender, voiceId, samplingRate |
Environment variable override: BHASHINI_TTS_MODEL_ID
Transliteration and Detection Service IDs
| Service ID | Capability | Description |
|---|---|---|
ai4bharat/indicxlit--gpu--t4 |
Transliteration | AI4Bharat IndicXlit neural transliteration across 22 Indic scripts |
ai4bharat/tld--gpu--t4 |
Language Detection | Text Language Detection across 22 Indic languages |
Environment variable overrides: BHASHINI_TRANSLITERATION_SERVICE_ID, BHASHINI_DETECTION_SERVICE_ID
Navana AI TTS
Navana Bodhi supports non-streaming and WebSocket TTS for Bengali (bn), English (en), Gujarati (gu), Hindi (hi), Kannada (kn), Malayalam (ml), Marathi (mr), Odia (or), Tamil (ta), and Telugu (te). Set NAVANA_API_KEY to enable the provider.
Voice IDs are shared across all ten languages. The default is default_female. The provider accepts voice, speed, num_step, g2p_overrides, and output_format for non-streaming requests. Live streaming accepts voice, num_step, and output_format; it does not accept speed. See Navana's voice reference for the latest account-specific voice list.
Microsoft Edge TTS
Microsoft Edge TTS provides high-quality neural voice synthesis for Indian languages without requiring an API key. Install via pip install 'indic-language-utils[tts-edge]'.
Neural Voice Identifiers
| Language | Language Code | Female Voice | Male Voice |
|---|---|---|---|
| Hindi | hi-IN / hi |
hi-IN-SwaraNeural (Default) |
hi-IN-MadhurNeural |
| English (India) | en-IN / en |
en-IN-NeerjaNeural |
en-IN-PrabhatNeural |
| Tamil | ta-IN / ta |
ta-IN-PallaviNeural |
ta-IN-ValluvarNeural |
| Telugu | te-IN / te |
te-IN-ShrutiNeural |
te-IN-MohanNeural |
| Bengali | bn-IN / bn |
bn-IN-TanishaaNeural |
bn-IN-BashkarNeural |
| Marathi | mr-IN / mr |
mr-IN-AarohiNeural |
mr-IN-ManoharNeural |
| Gujarati | gu-IN / gu |
gu-IN-DhwaniNeural |
gu-IN-NiranjanNeural |
| Kannada | kn-IN / kn |
kn-IN-SapnaNeural |
kn-IN-GaganNeural |
| Malayalam | ml-IN / ml |
ml-IN-SobhanaNeural |
ml-IN-MidhunNeural |
| Urdu (India) | ur-IN / ur |
ur-IN-GulNeural |
ur-IN-SalmanNeural |
| Nepali | ne-NP / ne |
ne-NP-HemkalaNeural |
ne-NP-SagarNeural |
Supported Parameters:
voice: Explicit voice identifier string (e.g.hi-IN-MadhurNeural).gender:"female"or"male"(automatically selects the default voice for the language).rate: Speed modifier string (e.g."+10%","-15%").pitch: Pitch modifier string (e.g."+2Hz","-2Hz").volume: Volume modifier string (e.g."+0%","-10%").
Environment variable override: EDGE_TTS_VOICE
Local and Offline Models
Faster-Whisper ASR
Faster-Whisper runs offline speech recognition locally on CPU or GPU. Install via pip install 'indic-language-utils[stt-whisper]'.
| Model Size | Parameters | VRAM (Approx) | Relative Speed |
|---|---|---|---|
tiny / tiny.en |
39 M | ~1 GB | Fastest |
base / base.en |
74 M | ~1 GB | Fast (Default) |
small / small.en |
244 M | ~2 GB | Balanced |
medium / medium.en |
769 M | ~5 GB | High Accuracy |
large-v3 |
1550 M | ~10 GB | Best Accuracy |
large-v3-turbo |
809 M | ~6 GB | Fast Large Model |
Environment variable overrides: FASTER_WHISPER_MODEL, WHISPER_MODEL
Aksharamukha Transliteration
Pure Python deterministic script converter supporting 120+ scripts and Romanization standards:
- Indic Scripts: Devanagari, Tamil, Telugu, Kannada, Malayalam, Bengali, Gujarati, Gurmukhi, Odia, Assamese, etc.
- Romanization Formats: IAST, ISO 15919, Harvard-Kyoto, Velthuis, SLP1, ITRANS, Titus, Roman (Colloquial).
Install via pip install 'indic-language-utils[local-transliteration]'.
FastText Language Detection
Offline compact neural text language detection model:
- Model ID:
lid.176.ftz(1 MB compressed) orlid.176.bin(126 MB full). - Identifies 176 languages and returns ranked probabilities.
Install via pip install 'indic-language-utils[local-tld]'.