Sarvam AI has released Saaras V4, the newest generation of its speech recognition model. It covers all 22 scheduled Indian languages plus English, now including global English accents. Sarvam reports state-of-the-art accuracy across all 22 languages.
Is it deployable? Yes, through Sarvam’s API today, using model="saaras:v4". Weights are not public, and Sarvam’s SageMaker self-hosting docs currently cover Saaras v3 only.
What is Inside Saaras V4
Saaras V4 is an encoder-decoder system. An audio encoder converts the waveform into embeddings that carry phonetic and acoustic detail. A temporal-downsampling adapter then shortens that sequence and projects it into the language model’s embedding space. This keeps long recordings inside the decoder’s context budget.
The decoder is Sarvam-3B, a 3B-parameter hybrid state-space language model trained from scratch in-house. It reads the audio features alongside a text prompt. It then emits the transcript autoregressively, feeding each token back as input for the next.
Benchmark Results
- English: Sarvam evaluated 7 English datasets. Six come from Hugging Face’s Open ASR Leaderboard: AMI, GigaSpeech, LibriSpeech clean, LibriSpeech other, SPGISpeech and VoxPopuli. The seventh is AI4Bharat’s Indian-accented Svarah. Scoring follows the leaderboard’s normalization code. Saaras V4 posts the lowest average WER among the models Sarvam benchmarked.
- Indic: On Vistaar, Sarvam reports results across 10 Indian languages using both WER and LLM-WER. LLM-WER adds a semantic check. It separates real meaning errors from harmless spelling or formatting variants common in Indic scripts.
- Noisy audio: On Kathbath Noisy, measured with LLM-WER, Sarvam says Saaras V4’s error rate is under half that of Deepgram Nova-3 and GPT-4o Transcribe. The set includes compressed, clipped and background-heavy recordings.
- Language ID: On verified IndicVoices utterances, language identification error is 2.9% across the top 10 Indian languages. It is 5.22% across all 22.
It is important to note that all numbers above are vendor-reported. Independent reproduction has not been published yet.
5 Output Modes From 1 Model
The same audio can return 5 representations, selected through the mode parameter:
- transcribe (default): native script with numbers and dates normalized.
- verbatim: every word as spoken, fillers and spoken numbers kept.
- codemix: native script, with English words left in English.
- translit: the full utterance in Latin script.
- translate: an English translation with numbers normalized.
Sarvam’s argument is simple. Handling these inside the model removes post-processing steps that can compound errors.
Keyterm Prompting
Keyterm prompting is new in V4 and works only with saaras:v4. You pass a JSON list under keyterms, with up to 50 terms of 64 characters each. Keyterms bias recognition; they do not guarantee output. Use codemix mode when a brand such as PhonePe must stay in Latin script.
On IndicContextEval (paper, Interspeech 2026), Saaras V4 reports 16.03% WER in the L5 keyword-prompting setting. Sarvam says that is the lowest score on the benchmark.
Streaming, Long Audio and Pricing
- Streaming: WebSocket with partial results and time to first token below 150 ms.
- REST: synchronous transcription for clips up to 30 seconds.
- Batch: asynchronous jobs up to 2 hours per file, with optional speaker diarization.
- SDKs: Python 3.9+ and Node.js 18+, plus LiveKit Agents, Pipecat and Vercel AI SDK integrations.
- Price: Sarvam lists speech-to-text at ₹30 per hour for real-time, streaming and batch, and ₹45 per hour with diarization.
Saaras v3 stays the default model. V4 uses the same request shape, so switching is a 1-line change.
‘+n.l+’
‘;d.onclick=function(){$$(‘.num’).forEach(function(x){x.classList.remove(‘on’)});d.classList.add(‘on’);$(‘#ninfo’).textContent=n.i};$(‘#nums’).appendChild(d)});
var done=false;
function animNums(){if(done)return;done=true;$$(‘.num’).forEach(function(d,i){var n=N[i],c=d.querySelector(‘.c’),t0=null,dec=(String(n.v).split(‘.’)[1]||”).length;
function f(ts){if(!t0)t0=ts;var k=Math.min((ts-t0)/1100,1),e=1-Math.pow(1-k,3);c.textContent=(n.v*e).toFixed(dec);if(k# pip install sarvamai\nfrom sarvamai import SarvamAI\nclient = SarvamAI(api_subscription_key=“YOUR_KEY”)\n\nresp = client.speech_to_text.transcribe(\n file=open(“audio.wav”, “rb”),\n model=“saaras:v4”,\n mode=“transcribe”,\n keyterms=[“Sarvam”, “New Delhi”],\n)\nprint(resp.transcript)’},
{t:”Long recordings”,s:”Calls, meetings, podcasts”,f:[“Batch API”,”Up to 2 hr per file”,”Diarization option”,”₹45/hr with diarization”],c:’job = client.speech_to_text_job.create_job(\n model=“saaras:v4”,\n mode=“transcribe”,\n language_code=“hi-IN”,\n with_diarization=True,\n keyterms=[“Sarvam”, “Vistaar”],\n)\njob.upload_files(file_paths=[“interview.wav”])\njob.start()\njob.wait_until_complete()\njob.download_outputs(output_dir=“./output”)’},
{t:”Voice agents”,s:”Live, low latency”,f:[“WebSocket”,”TTFT under 150 ms”,”LiveKit / Pipecat”,”Keyterms via URL param”],c:’# keyterms go in the connection URL (SDK connect() does not expose them yet)\nparams = urllib.parse.urlencode({\n “model”: “saaras:v4”,\n “language_code”: “hi-IN”,\n “keyterms”: json.dumps([“Sarvam”, “New Delhi”]),\n})\nurl = f”wss://api.sarvam.ai/speech-to-text/ws?{params}”‘}];
function dep(i){$$(‘.q’).forEach(function(q,k){q.classList.toggle(‘on’,k===i)});$(‘#facts’).innerHTML=D[i].f.map(function(x){return ‘‘+x+’‘}).join(”)+’₹30/hr base‘;$(‘#code’).innerHTML=D[i].c;setTimeout(resize,60)}
D.forEach(function(d,i){var b=document.createElement(‘button’);b.className=”q”;b.innerHTML=’‘+d.t+’‘+d.s+’‘;b.onclick=function(){dep(i)};$(‘#qs’).appendChild(b)});
dep(0);
window.addEventListener(‘load’,resize);window.addEventListener(‘resize’,resize);setTimeout(resize,300);
})();
Credit: Source link
