 |
VOOZH | about |
|
VOOZH | about |
This page provides a comprehensive overview of the speech recognition services supported by the SpeechRecognition library. It covers both online API-based services and offline recognition engines, their integration within the library, and how to use them. For information about the core architecture of the library, see Core Architecture.
The SpeechRecognition library supports two main categories of speech recognition services:
Online Services - These require an internet connection and typically offer high accuracy but may have usage limitations or costs:
Offline Engines - These run locally without an internet connection, providing privacy and consistent availability:
The diagram below illustrates how the various speech recognition services integrate with the library:
Sources: speech_recognition/__init__.py603-1600 README.rst28-43
The following table provides details on all supported speech recognition services:
| Service | Type | Method | Authentication Required | Key/Token Format |
|---|---|---|---|---|
| Google Speech | Online | recognize_google | Optional | API Key (string) |
| Google Cloud | Online | recognize_google_cloud | Required | Google Application Credentials |
| Wit.ai | Online | recognize_wit | Required | 32-char alphanumeric string |
| Azure Speech | Online | recognize_azure | Required | 32-char hexadecimal string |
| Bing Voice (Deprecated) | Online | recognize_bing | Required | 32-char hexadecimal string |
| Houndify | Online | recognize_houndify | Required | Client ID and Key (Base64) |
| IBM Speech to Text | Online | recognize_ibm | Required | Username and Password |
| OpenAI Whisper API | Online | recognize_openai | Required | API Key via env variable |
| Groq Whisper API | Online | recognize_groq | Required | API Key via env variable |
| CMU Sphinx | Offline | recognize_sphinx | No | None |
| Whisper | Offline | recognize_whisper | No | None |
| Faster Whisper | Offline | recognize_faster_whisper | No | None |
Sources: speech_recognition/__init__.py603-1600 README.rst28-43
This diagram illustrates the typical flow of audio data through the recognition process:
Sources: speech_recognition/__init__.py318-602 examples/microphone_recognition.py9-104 examples/audio_transcribe.py1-87
The standard Google Speech Recognition service provides good recognition capabilities without requiring an API key, though one can be provided.
The API limits usage to 50 requests per day without a key. With a key, higher limits apply.
Sources: speech_recognition/__init__.py640-710 examples/microphone_recognition.py23-32
Google Cloud Speech provides enterprise-level recognition with higher accuracy and more features than the standard Google service.
Requires Google Cloud credentials with the Speech-to-Text API enabled.
Sources: speech_recognition/__init__.py711-864 examples/microphone_recognition.py34-41 examples/special_recognizer_features.py34-43
Wit.ai offers speech recognition with natural language understanding capabilities.
Requires a Wit.ai account and API key.
Sources: speech_recognition/__init__.py603-638 examples/microphone_recognition.py43-50
Azure Speech provides cloud-based speech recognition for a variety of languages.
Requires an Azure Speech subscription key.
Sources: speech_recognition/__init__.py964-1052 examples/microphone_recognition.py61-68
This service is now deprecated but still supported for backward compatibility.
Sources: speech_recognition/__init__.py865-963 examples/microphone_recognition.py52-59
Houndify provides speech recognition with natural language understanding and domain-specific functionality.
Requires Houndify client ID and key.
Sources: speech_recognition/__init__.py1053-1141 examples/microphone_recognition.py70-78
IBM's Watson Speech to Text service provides enterprise-level speech recognition.
Requires IBM Cloud credentials.
Sources: speech_recognition/__init__.py1142-1298 examples/microphone_recognition.py80-88
Integration with OpenAI's cloud-based Whisper model.
Requires the OpenAI API key to be set as an environment variable.
Sources: speech_recognition/__init__.py1368-1416 examples/microphone_recognition.py98-104
Integration with Groq's implementation of the Whisper model.
Requires the Groq API key to be set as an environment variable.
Sources: speech_recognition/__init__.py1417-1465
Sphinx is an offline speech recognition engine that doesn't require internet connectivity.
Supports keyword recognition and grammar-based recognition.
Sources: speech_recognition/__init__.py1299-1367 examples/special_recognizer_features.py14-31
Local implementation of OpenAI's Whisper model running entirely on the user's device.
Requires the Whisper package to be installed.
Sources: speech_recognition/__init__.py1466-1516 examples/microphone_recognition.py90-96
An optimized implementation of Whisper for faster performance on local hardware.
Requires the faster-whisper package to be installed.
Sources: speech_recognition/__init__.py1517-1579
The diagram below shows how the Recognizer class integrates with various speech recognition services and the transformation of audio data throughout the process:
Sources: speech_recognition/__init__.py318-1600
Most online services require authentication. The following table summarizes the authentication methods for each service:
| Service | Environment Variable | Method Parameter | Authentication Format |
|---|---|---|---|
| Google Speech | N/A | key | API Key (string) |
| Google Cloud | N/A | credentials_json | Path to JSON or JSON content |
| Wit.ai | N/A | key | API Key (string) |
| Azure Speech | N/A | key | API Key (string) |
| Bing Voice | N/A | key | API Key (string) |
| Houndify | N/A | client_id, client_key | Client ID and key (strings) |
| IBM Speech | N/A | username, password | Service credentials |
| OpenAI Whisper | OPENAI_API_KEY | N/A | API Key (string) |
| Groq Whisper | GROQ_API_KEY | N/A | API Key (string) |
Sources: speech_recognition/__init__.py603-1465 README.rst152-202
Most services support multiple languages. Language can be specified using language codes:
Each service uses slightly different language code formats. Common formats include:
Sources: speech_recognition/__init__.py640-1579 README.rst223-229
All recognition methods support a show_all parameter that returns the complete service response rather than just the recognized text:
This is useful for accessing additional information such as:
Sources: examples/extended_results.py1-87
The SpeechRecognition library provides a unified interface to multiple speech recognition services, both online and offline. Each service has its own strengths, language support, and authentication requirements. The choice of service depends on specific needs such as accuracy requirements, privacy concerns, internet connectivity, and language support.
The library's design makes it easy to switch between services or to use multiple services for redundancy or comparison purposes.
Sources: speech_recognition/__init__.py1-1600 README.rst24-44
Refresh this wiki