Tasks and inputs
Nemotron-Labs-Unified-ASR-0.6b is listed for Automatic speech recognition, Speech → Text. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
Nemotron-Labs-Unified-ASR-0.6b is NVIDIA's open 600M speech recognizer for 25 European languages, running offline or streaming at down to 160 ms.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights; no per-token price is published. Your cost is the GPU time you run it on.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Teams shipping live captions usually keep two speech models: a streaming one for latency and an offline one for accuracy. Nemotron-Labs-Unified-ASR-0.6b collapses them into a single checkpoint. It is the multilingual version of NVIDIA's English Parakeet unified model, covering 25 European languages, and it runs in either mode with a minimum algorithmic latency of 160 ms.
The network is a 600M-parameter FastConformer encoder (24 layers) with an RNN-Transducer decoder. During training it saw both full-context attention and chunked attention masks, plus a consistency loss that narrows the gap between modes. All parameters are shared. The encoder was initialised from parakeet-tdt-0.6b-v3 and trained mainly on the Granary dataset.
Language choice is a prompt. You pass a language ID to the encoder, which steadies output at very low latency where models tend to drift between languages. An automatic mode, target_lang=auto, is there when you do not know.
Latency is a dial. Chunk plus right context can be set from 2,080 ms down to 160 ms in 80 ms steps, with no retraining.
There is a dating quirk. The card gives 27 October 2026 as the release, but the repository has existed since 21 August. Treat late October as the public date.
Record a few minutes of your real audio — call-centre lines, meeting rooms, whatever you serve — in two or three of the languages that matter. Transcribe it offline, then at 1,040 ms, 480 ms and 160 ms. Plot word error rate against latency and choose the knee of the curve. If you feed an LLM translator downstream, keep a little right context; it lets the current chunk peek at upcoming audio before the text is committed.
For an English-centric hosted option, compare Whisper large-v3 Turbo. More options sit under speech recognition.
Compare the model on the work you need to complete.
Nemotron-Labs-Unified-ASR-0.6b is listed for Automatic speech recognition, Speech → Text. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under OpenMDW-1.1. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
assemblyai
AssemblyAI-STT-3.7 is AssemblyAI's November 2026 pre-recorded transcription default: 32 languages, new diarization, fewer language flips, $0.21 per hour.
assemblyai
AssemblyAI-STT-3.7-Realtime streams transcripts in 32 languages for captions, notetakers and agent assist, with final speaker labels per turn at $0.30/hour.
assemblyai
AssemblyAI-STT-3.7-Realtime-Agent is a streaming recogniser for voice agents and IVR: 32 languages, 86 ms to final transcript, 4.46% vendor WER, $0.45/hour.
salesforce
SOV-Transcribe-1 is Salesforce's open-weight speech-augmented model for non-streaming transcription, with keyword biasing and style instructions; Apache-2.0.
ElevenLabs
Scribe v3 Live is ElevenLabs' streaming transcription model with turn prediction, mid-sentence language switching and a 10,000-term keyterm library.
NVIDIA
Nemotron 3.5 Transcribe 30B A4B is NVIDIA's open offline ASR model: 40 languages, hour-long files, word timestamps, diarisation and translation.