DeutscheKI/tevr-asr-tool
A 284-line C++ CLI tool performing state-of-the-art German speech-to-text using wav2vec2 acoustic models and KenLM language models.
Velocity · 7d
+0.3
★ / day
Trend
→steady
star history
This tool provides offline, private German speech recognition by running a pre-trained wav2vec2-xls-r-1b acoustic model with a KenLM language model for beam search re-scoring. Users pass a WAV file to the CLI and receive transcribed text. The acoustic model is sourced from HuggingFace transformers, while the language model handles token logits to text conversion and scoring.