“Sorry, Could you repeat that again?” - Speech Recognition with Python
summary
A working Python pipeline converts audio to MFCC features, trains a hidden Markov model + neural hybrid acoustic model and an n‑gram or neural language model, and runs beam search recognition with alignments and RTF reporting. It uses datasets like Google Speech Commands, LibriSpeech and Common Voice, interfaces to TensorFlow or PyTorch, and a toolkit referenced as get the trans lectures you pivot toolkit / PI tlk. Example evaluation shows about 11.5% WER on a sample and produces time alignments for segmentation.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:01What is this talk about?
- 1:26Some remarks...
- 1:48Automatic Speech Recognition (ASR)
- 2:00ASR is not already solved?
- 3:06Technical Challenges
- 3:23The big picture
- 3:50ASR System: Steps
- 4:12ASR System: Toolkits
- 5:17ASR System: Open Resources
- 6:00ASR System: Prepare resources
- 7:07ASR System: Resources
- 8:12ASR System: Feature Extraction
- 9:50ASR System: Modelling
- 10:03ASR System: Acoustic Modelling
- 15:17ASR System: Language Modelling
- 20:33ASR System: Recognition graph
- 26:52ASR System: Evaluation
- 28:41ASR System: Advanced Topics
- 30:05What was this talk about?
related talks
29mRecognizing Speech in Python
25mBeyond Transcription: Building Voice AI That Understands Conversations
40mGathering Insights from Audio Data
26mMake Python Talk, Make Python Listen
23mMultilingual Natural Language Processing
36mNVIDIA NeMo: Toolkit for Conversational AI
16mThe End of Awkward AI Transcriptions
13mMining Parliamentary Gold - Building Hebrew ASR