Using the API#

The Google Speech API enables developers to convert audio to text. The API recognizes over 80 languages and variants, to support your global user base.

Warning

This is a Beta release of Google Speech API. This API is not intended for real-time usage in critical applications.

Client#

Client objects provide a means to configure your application. Each instance holds an authenticated connection to the Natural Language service.

For an overview of authentication in google-cloud-python, see Authentication.

Assuming your environment is set up as described in that document, create an instance of Client.

>>> from google.cloud import speech
>>> client = speech.Client()

Asychronous Recognition#

The async_recognize() sends audio data to the Speech API and initiates a Long Running Operation. Using this operation, you can periodically poll for recognition results. Use asynchronous requests for audio data of any duration up to 80 minutes.

Note

Only the Encoding.LINEAR16 encoding type is supported by asynchronous recognition.

See: Speech Asynchronous Recognize

>>> import time
>>> from google.cloud import speech
>>> client = speech.Client()
>>> sample = client.sample(source_uri='gs://my-bucket/recording.flac',
...                        encoding=speech.Encoding.LINEAR16,
...                        sample_rate=44100)
>>> operation = sample.async_recognize(max_alternatives=2)
>>> retry_count = 100
>>> while retry_count > 0 and not operation.complete:
...     retry_count -= 1
...     time.sleep(10)
...     operation.poll()  # API call
>>> operation.complete
True
>>> for result in operation.results:
...     for alternative in result.alternatives:
...         print('=' * 20)
...         print(alternative.transcript)
...         print(alternative.confidence)
====================
'how old is the Brooklyn Bridge'
0.98267895

Synchronous Recognition#

The sync_recognize() method converts speech data to text and returns alternative text transcriptons.

This example uses language_code='en-GB' to better recognize a dialect from Great Britian.

>>> from google.cloud import speech
>>> client = speech.Client()
>>> sample = client.sample(source_uri='gs://my-bucket/recording.flac',
...                        encoding=speech.Encoding.FLAC,
...                        sample_rate=44100)
>>> results = sample.sync_recognize(
...     speech.Encoding.FLAC, 16000,
...     source_uri='gs://my-bucket/recording.flac', language_code='en-GB',
...     max_alternatives=2)
>>> for result in results:
...     for alternative in result.alternatives:
...         print('=' * 20)
...         print('transcript: ' + alternative.transcript)
...         print('confidence: ' + alternative.confidence)
====================
transcript: Hello, this is a test
confidence: 0.81
====================
transcript: Hello, this is one test
confidence: 0

Example of using the profanity filter.

>>> from google.cloud import speech
>>> client = speech.Client()
>>> sample = client.sample(source_uri='gs://my-bucket/recording.flac',
...                        encoding=speech.Encoding.FLAC,
...                        sample_rate=44100)
>>> results = sample.sync_recognize(max_alternatives=1,
...                                 profanity_filter=True)
>>> for result in results:
...     for alternative in result.alternatives:
...         print('=' * 20)
...         print('transcript: ' + alternative.transcript)
...         print('confidence: ' + alternative.confidence)
====================
transcript: Hello, this is a f****** test
confidence: 0.81

Using speech context hints to get better results. This can be used to improve the accuracy for specific words and phrases. This can also be used to add new words to the vocabulary of the recognizer.

>>> from google.cloud import speech
>>> client = speech.Client()
>>> sample = client.sample(source_uri='gs://my-bucket/recording.flac',
...                        encoding=speech.Encoding.FLAC,
...                        sample_rate=44100)
>>> hints = ['hi', 'good afternoon']
>>> results = sample.sync_recognize(max_alternatives=2,
...                                 speech_context=hints)
>>> for result in results:
...     for alternative in result.alternatives:
...         print('=' * 20)
...         print('transcript: ' + alternative.transcript)
...         print('confidence: ' + alternative.confidence)
====================
transcript: Hello, this is a test
confidence: 0.81

Streaming Recognition#

The streaming_recognize() method converts speech data to possible text alternatives on the fly.

Note

Streaming recognition requests are limited to 1 minute of audio.

See: https://cloud.google.com/speech/limits#content

>>> from google.cloud import speech
>>> client = speech.Client()
>>> with open('./hello.wav', 'rb') as stream:
...     sample = client.sample(stream=stream,
...                            encoding=speech.Encoding.LINEAR16,
...                            sample_rate=16000)
...     results = list(sample.streaming_recognize())
>>> print(results[0].alternatives[0].transcript)
'hello'
>>> print(results[0].alternatives[0].confidence)
0.973458576

By default the API will perform continuous recognition (continuing to process audio even if the speaker in the audio pauses speaking) until the client closes the output stream or until the maximum time limit has been reached.

If you only want to recognize a single utterance you can set
single_utterance to True and only one result will be returned.

See: Single Utterance

>>> with open('./hello_pause_goodbye.wav', 'rb') as stream:
...     sample = client.sample(stream=stream,
...                            encoding=speech.Encoding.LINEAR16,
...                            sample_rate=16000)
...     responses = sample.streaming_recognize(single_utterance=True)
...     results = list(responses)
>>> print(results[0].alternatives[0].transcript)
hello
>>> print(results[0].alternatives[0].confidence)
0.96523453546

If interim_results is set to True, interim results (tentative hypotheses) may be returned as they become available.

>>> from google.cloud import speech
>>> client = speech.Client()
>>> with open('./hello.wav', 'rb') as stream:
...     sample = client.sample(stream=stream,
...                            encoding=speech.Encoding.LINEAR16,
...                            sample_rate=16000)
...     for results in sample.streaming_recognize(interim_results=True):
...         print('=' * 20)
...         print(results[0].alternatives[0].transcript)
...         print(results[0].alternatives[0].confidence)
...         print(results[0].is_final)
...         print(results[0].stability)
====================
'he'
None
False
0.113245
====================
'hell'
None
False
0.132454
====================
'hello'
0.973458576
True
0.982345