Recognizing Speech in Python
summary
A complete Python recognizer extracts 13 MFCC coefficients every 10 ms into NumPy arrays, trains on tiny datasets (15/20 cat or dog clips, 50 digit samples) and classifies by distance to cluster mean — everything under ~500 lines. Uses SciPy/NumPy for feature extraction, hits Google API limits (50 requests/day) and language gaps, and shows PocketSphinx for constrained grammars; notes practical problems: per-speaker models, end-pointing, background noise, and ~75% lab accuracy for small open models.
AI-written summary · about
chapters (11)
- 0:00Speech recognition in Python
- 1:44Limitations of speech APIs
- 2:48Custom models for non-human sounds
- 4:13Training animal sound model
- 7:55Testing animal classifier
- 12:02Recognizing spoken digits
- 15:41Acoustic and language models
- 18:08Finite state grammars
- 20:12Customizing Pocketsphinx
- 24:12Real-world speech challenges
- 26:43Q&A
related talks
31m“Sorry, Could you repeat that again?” - Speech Recognition with Python
31mClassifying Audio into Types using Python
40mBuilding Features and Machine Learning Models for Audio Signals
40mGathering Insights from Audio Data
11mClassifying Spoken Location Names
37mSpeech to text with Elasticsearch
22mImplementation and analysis of deep learning models for speech classification
25mLooking at Sound