code-switch

June 26, 2024 ยท View on GitHub

Detailed documentations can be found at /docs

Setup

  1. Install external dependencies if your system do no have them.
  2. Create a conda environment and activate it
  3. Install python 3.11 in your conda environment.
  4. Run make

External dependencies:

  • libboost

Run linters

Run make lint

Run formatters

Run make lint-format

Results:

Model NameTest Dataset NameWERCER
Nemo stt_enes_conformer_transducer_large_codesw beam width 16Miami eng herring187.77%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + RedPajamaV2 KenLMBark TTS Synthetic 2024060529.37%16.71%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + synthetic KenLMBark TTS Synthetic 2024060524.02%14.04%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + synthetic KenLMCommonvoice es en dev15.65%8.92%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + redpajama KenLMCommonvoice es en dev13.62%7.65%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + Commonvoice KenLMCommonvoice es en dev13.62%7.65%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + synthetic 50k KenLMBark TTS Synthetic 2024060525.34%14.76%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16Bark TTS Synthetic 2024061929.78%13.88%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16Commonvoice es en dev12.38%4.85%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + synthetic 50k KenLMBark TTS Synthetic 2024061926.63%17.09%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + synthetic 50k KenLMCommonvoice es en dev15.67 %8.82%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + commonvoice lm mix synthetic 50k KenLMBark TTS Synthetic 2024061927.70%18.06%
Nemo stt_enes_conformer_transducer_large_codesw beam width 16 + commonvoice lm mix synthetic 50k KenLMCommonvoice es en dev14.28%8.05%