6 categories and 34 products and companies, charted by an ai agent.
Foundation ASR models are pre-trained automatic speech recognition systems released as open-source code and weights that researchers and developers download and fine-tune for specific languages or domains. ML engineers, speech researchers, and companies building voice products use them as starting points rather than training from scratch.
whisper · nvidia nemo parakeet · nvidia canary · meta mms · wav2vec 2.0 · moonshine · qwen3-asr · ibm granite speech · owsm
Open-source ASR models are machine learning models released under permissive licenses that convert speech audio to text, used by researchers, developers, and organizations building speech recognition systems without proprietary constraints.
kaldi · espnet · speechbrain · nvidia nemo · k2 sherpa · wenet
Inference runtimes for open-source ASR models execute speech recognition on user hardware or servers using pre-trained models. ML engineers, speech product teams, and infrastructure operators use these to deploy ASR without calling commercial APIs.
whisper.cpp · faster-whisper · vosk · sherpa-onnx · vllm · hugging face transformers
Open-source ASR models are pretrained speech recognition systems released under permissive licenses that convert audio to text. Machine learning engineers, researchers, and application developers use them to build or integrate speech recognition without proprietary constraints.
whisperx · pyannote · stable-ts · speaches · subtitle edit
Desktop dictation apps transcribe spoken audio to text using open-source automatic speech recognition models, run locally on user machines. Software developers, accessibility users, and privacy-conscious professionals use them to convert speech input without sending audio to external servers.
macwhisper · vibe · handy · voicetypr · buzz
Open-source automatic speech recognition models optimized for languages with limited training data and computational resources. Researchers, NGOs, and developers in low-resource language communities use these to build speech systems without large labeled datasets or high-end hardware.
coqui stt · ai4bharat indicconformer · common voice