Skip to content

Improved speech-to-text module

  • Migration to Faster-Whisper after benchmarking:
  • Improved speed (halved translation time).
  • Higher accuracy in noisy environments.
  • Dynamic Integration of "Hot Words":
  • Context-specific vocabulary dynamically adjusted.
  • Increases robustness and accuracy for uncommon terms.

STT Benchmark

File (10s) Size (MB) Faster-whisper accuracy Time (s) Whisper accuracy Time (s)
test1.wav 1.22 85.7% 0.64 71.4% 1.25
test2.wav 1.22 77.8% 0.71 33.3% 1.44
test3.wav 1.22 71.4% 0.66 57.1% 1.13
test4.wav 1.22 80% 0.70 60% 1.36
test5.wav 1.53 71.4% 4.68 71.4% 4.5
test6.wav 1.83 42.9% 0.63 28.6% 1.03
test7.wav 1.83 90% 0.64 90% 0.87
test8.wav 1.83 83.3% 0.61 66.7% 0.99
test9.wav 1.83 100% 0.62 100% 0.94
test10.wav 1.83 100% 0.58 100% 0.77