nobody132/masr

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1968 stars · 477 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MASR is an end-to-end Mandarin Chinese automatic speech recognition project built on a gated convolutional neural network (similar to Wav2Letter but using GLU activations) implemented in PyTorch. It includes pretrained models trained on the AISHELL-1 dataset (~150 hours) and supports optional external language models to reduce character error rate.

Use cases

• transcribe Mandarin Chinese audio to text
• train a Chinese speech recognition model on AISHELL-1
• experiment with convolutional neural networks for ASR
• fine-tune or evaluate a Mandarin ASR model by CER
• integrate Chinese voice input into a Python application
• study how GLU activations affect ASR convergence

When to choose

• you need offline Mandarin speech recognition in Python with PyTorch
• you want a small, readable reference implementation of convolutional ASR for Chinese
• you want pretrained Chinese ASR models without heavy infrastructure
• you are researching gated convolutional networks for speech recognition

When to avoid

• you need production-grade accuracy comparable to industrial ASR systems
• you need multilingual or non-Mandarin speech recognition
• you want a maintained project with a clear license for commercial use
• you need streaming/real-time low-latency ASR out of the box

Facets

library · maturity maintenance

speech-recognition machine-learning deep-learning audio-processing speech-processing machine-learning python mandarin chinese-asr pytorch wav2letter gated-convolutional-network aishell-1 pretrained-models natural-language-processing linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="nobody132/masr")