Google Cloud Speech Tool

This is a quick hack script made in a few days, but I hope to use it as a guide in building other tools that work with the Google Cloud Speech API.

Getting started

Install the dependencies. Note that this script uses ffmpy which requires ffmpeg to be installed on the system.

pip install -r requirements.txt

The configuration file encapsulates a few options used by the script.

Name	Description
google_cloud (dict)	GC specific settings
google_cloud.credentials_file (str)	simple path to a GC project credentials file
google_cloud.storage (dict)	details related to GC Storage
google_cloud.storage.bucket (str)	bucket name where to store files in case you want to upload audio files and analyze them from GCS
google_cloud.storage.remote_dir (str)	sub directory within the bucket where the script can freely upload/destroy files

The script a2t.py provides four main actions:

• --action-upload: uploads an audio file to GCS
• --action-force: if the current audio file (-a) has been previously converted and exists on the FS or in GCS, destroy both of them
• --action-compat: runs an ffmpeg command to convert the input file (-a) into a single channel with the format set in --file-ext
• --action-text: submits the compatible audio file created by --action-compat to the speech recognition API

The remainder of the options apply to the speech recognition API:

• --sample-rate: indicates the sample rate of the input file
• --sample-enc: indicates the sample encoding (recommend --async --sample-enc LINEAR16, --sample-enc FLAC # without --async) (see google.cloud.speech.Encoding.* for valid all values)
• --file-ext: indicates the file extension of the desired format for upload (recommend --async --file-ext .wav, --file-ext .flac # without --async)
• --async: calls the async_recognize Speech API function; calls sync_recognize if this is not present
• --async-retry: indicates the number of times to call and check for the result of an async_recognize operation
• --async-sleep: indicates the number of seconds to wait in between checks of an async_recognize operation
• --max-alts: indicates the number of alternatives that will be returned from the Speech API
• --interval: this extracts an interval of the input audio file for conversion to text (i.e. 5,10 indicates from 5s to 10s)
• --lang-code: this indicates the language code to pass to the Speech API as a hint

Usage

usage: a2t.py [-h] -c CONFIG_FILENAME -a AUDIO_FILENAME [-q] [--action-force]
              [--action-compat] [--action-upload] [--action-text]
              [--sample-rate SAMPLE_RATE] [--sample-enc SAMPLE_ENCODING]
              [--interval INTERVAL] [--lang-code LANGUAGE_CODE]
              [--file-ext FILE_EXT] [--async] [--async-sleep ASYNC_SLEEP]
              [--async-retry ASYNC_RETRY] [--max-alts MAX_ALTS]

Example

#!/bin/bash
#
# This example illustrates a working set of options for the async recognition method
#
FILENAME=$1
[[ -z "$FILENAME" || ! -f "$FILENAME" ]] && echo "audio file not found" && exit 1

python gcs_tool/a2t.py -c config.yml -a "$FILENAME" \
  --interval 0,59 \
  --sample-enc LINEAR16 \
  --file-ext .wav \
  --async --async-sleep 2 --async-retry 150 \
  --max-alts 1 \
  --lang-code en-GB \
  --action-compat \
  --action-upload \
  --action-text \
  --action-force