user@devops:~$ cat README.md
Whisper STT
# Description
Speech recognition project with OpenAI Whisper, an encoder-decoder transformer model trained on 680,000 hours of multilingual data. The pipeline generates real audio with gTTS in Spanish, English and bilingual, and transcribes with tiny (37M params, ~142MB) and base (72M params, ~274MB) models. Includes detailed WER (Word Error Rate), CER (Character Error Rate), RTF (Real-Time Factor) comparison, automatic language detection (99.9% accuracy for Spanish and English), word-level timestamps with per-word confidence, and export to SRT, VTT, TSV and JSON formats. The base model achieves 0% WER in English and 7.4% in Spanish on clear audio.
# Key features
$ Multi-language transcription with OpenAI Whisper (tiny + base, 37M–72M parameters)
$ Word Error Rate (WER) and Character Error Rate (CER) as accuracy metrics
$ Automatic language detection with 99.9% accuracy (Spanish, English)
$ Word-level timestamps with per-word confidence for exact alignment
$ Export to professional formats: SRT, VTT, TSV, JSON
$ 6 visualizations: waveform, model comparison, timeline segments, language matrix
$ Real audio synthesized with gTTS (Google Text-to-Speech) in 3 languages
# Gallery
# Technologies used
