Google Speech-to-Text Audio Transcripts

The GoogleSpeechToTextLoader allows to transcribe audio files with the Google Cloud Speech-to-Text API and loads the transcribed text into documents.

To use it, you should have the google-cloud-speech python package installed, and a Google Cloud project with the Speech-to-Text API enabled.

• Bringing the power of large models to Google Cloud’s Speech API

Installation & setup

First, you need to install the google-cloud-speech python package.

You can find more info about it on the Speech-to-Text client libraries page.

Follow the quickstart guide in the Google Cloud documentation to create a project and enable the API.

%pip install --upgrade --quiet langchain-google-community[speech]

Example

The GoogleSpeechToTextLoader must include the project_id and file_path arguments. Audio files can be specified as a Google Cloud Storage URI (gs://...) or a local file path.

Only synchronous requests are supported by the loader, which has a limit of 60 seconds or 10MB per audio file.

from langchain_google_community import GoogleSpeechToTextLoader

project_id = "<PROJECT_ID>"
file_path = "gs://cloud-samples-data/speech/audio.flac"
# or a local file path: file_path = "./audio.wav"

loader = GoogleSpeechToTextLoader(project_id=project_id, file_path=file_path)

docs = loader.load()

Note: Calling loader.load() blocks until the transcription is finished.

The transcribed text is available in the page_content:

docs[0].page_content

"How old is the Brooklyn Bridge?"

The metadata contains the full JSON response with more meta information:

docs[0].metadata

{
  'language_code': 'en-US',
  'result_end_offset': datetime.timedelta(seconds=1)
}

Recognition Config

You can specify the config argument to use different speech recognition models and enable specific features.

Refer to the Speech-to-Text recognizers documentation and the RecognizeRequest API reference for information on how to set a custom configuation.

If you don't specify a config, the following options will be selected automatically:

• Model: Chirp Universal Speech Model
• Language: en-US
• Audio Encoding: Automatically Detected
• Automatic Punctuation: Enabled

from google.cloud.speech_v2 import (
    AutoDetectDecodingConfig,
    RecognitionConfig,
    RecognitionFeatures,
)
from langchain_google_community import GoogleSpeechToTextLoader

project_id = "<PROJECT_ID>"
location = "global"
recognizer_id = "<RECOGNIZER_ID>"
file_path = "./audio.wav"

config = RecognitionConfig(
    auto_decoding_config=AutoDetectDecodingConfig(),
    language_codes=["en-US"],
    model="long",
    features=RecognitionFeatures(
        enable_automatic_punctuation=False,
        profanity_filter=True,
        enable_spoken_punctuation=True,
        enable_spoken_emojis=True,
    ),
)

loader = GoogleSpeechToTextLoader(
    project_id=project_id,
    location=location,
    recognizer_id=recognizer_id,
    file_path=file_path,
    config=config,
)

Related

• Document loader conceptual guide
• Document loader how-to guides