 |
VOOZH | about |
|
VOOZH | about |
This page documents the advanced features and customization options available in the SpeechRecognition library beyond the basic recognition functionality. These features provide greater control over recognition behavior, audio processing, and continuous operation.
For basic recognition patterns, see Usage Patterns. For specific information about energy threshold adjustment, see Energy Threshold Adjustment. For details on audio manipulation, see Audio Manipulation.
The SpeechRecognition library includes sophisticated mechanisms to dynamically adjust the energy threshold that determines when speech begins and ends.
The Recognizer class provides the following properties to control energy threshold behavior:
| Property | Default | Description |
|---|---|---|
energy_threshold | 300 | Minimum audio energy level to consider as speech |
dynamic_energy_threshold | True | Whether to automatically adjust threshold |
dynamic_energy_adjustment_damping | 0.15 | How quickly to adjust the threshold (lower = faster) |
dynamic_energy_ratio | 1.5 | Target energy ratio for adjustment |
When dynamic_energy_threshold is enabled, the library continuously adapts the threshold based on ambient noise levels. This enables more accurate speech detection across different environments.
Sources: speech_recognition/__init__.py323-326 speech_recognition/__init__.py502-506
The adjust_for_ambient_noise() method calibrates the energy threshold to account for background noise, improving recognition in noisy environments.
Example usage:
The duration parameter controls how long the calibration runs. Longer durations produce more accurate calibration but add delay before recognition can begin.
Sources: speech_recognition/__init__.py366-391 examples/calibrate_energy_threshold.py
One of the most powerful advanced features is the ability to continuously listen for speech in the background while your application performs other tasks.
The listen_in_background() method:
Example usage:
Sources: speech_recognition/__init__.py570-601 examples/background_listening.py
The library supports various advanced options for speech recognition engines that can significantly improve accuracy and control.
All recognition methods support a show_all parameter that returns the complete recognition data instead of just the top result:
This provides access to alternative transcriptions, confidence scores, and other service-specific data, which can be valuable for implementing more sophisticated applications.
Sources: speech_recognition/__init__.py635-638 examples/extended_results.py
The Sphinx recognizer (offline) offers specialized control through:
Example with keyword entries:
Example with grammar file:
Sources: speech_recognition/__init__.py examples/special_recognizer_features.py14-31 examples/counting.gram
Some services, like Google Cloud Speech, support preferred phrases to improve recognition of specific terms:
This is particularly useful for domain-specific terminology or proper nouns that might otherwise be misrecognized.
Sources: examples/special_recognizer_features.py34-43
The library integrates with Snowboy for offline hotword detection, enabling wake-word functionality similar to commercial voice assistants.
Example usage:
This feature is particularly useful for always-on applications that should only process speech after a specific trigger phrase.
Sources: speech_recognition/__init__.py393-440 speech_recognition/__init__.py442-457
The library provides sophisticated audio handling capabilities for working with various formats and processing requirements.
The library handles conversions between various audio formats transparently:
The AudioData class methods handle complex audio transformations:
| Method | Parameters | Description |
|---|---|---|
get_wav_data() | convert_rate, convert_width | Get WAV-encoded audio with optional conversion |
get_flac_data() | convert_rate, convert_width | Get FLAC-encoded audio with optional conversion |
get_raw_data() | convert_rate, convert_width | Get raw audio data with optional conversion |
The library automatically handles:
Sources: speech_recognition/__init__.py272-315
The Microphone class offers advanced configuration options for audio capture:
Utility methods are provided to help find and select the correct microphone:
Sources: speech_recognition/__init__.py69-163
The Recognizer provides fine-grained control over how phrases are detected and processed:
| Property | Default | Description |
|---|---|---|
pause_threshold | 0.8 | Seconds of non-speaking audio to mark end of phrase |
phrase_threshold | 0.3 | Minimum seconds of speaking audio to consider as a phrase |
non_speaking_duration | 0.5 | Non-speaking audio to keep on both sides of recording |
operation_timeout | None | Seconds to wait for API responses before timeout |
These properties allow customization of the recognition behavior to better match different speech patterns and use cases.
Examples of adjusting these parameters:
Sources: speech_recognition/__init__.py323-331 speech_recognition/__init__.py474-476
The listen() method supports a streaming mode that yields audio data chunks as they become available:
This enables real-time processing of speech as it's being spoken, rather than waiting for the complete phrase to be finished.
Sources: speech_recognition/__init__.py442-568
The library's architecture makes it easy to fallback between different recognition engines based on needs:
This allows implementing robust fallback strategies:
Sources: speech_recognition/__init__.py603-4102 README.rst28-43
Refresh this wiki