speechmatics-python  

Python client library and CLI for Speechmatics Realtime and Batch ASR v2 APIs.

Getting started

To install from PyPI:

pip install speechmatics-python

To install from source:

git clone https://github.com/speechmatics/speechmatics-python
cd speechmatics-python && python setup.py install

Windows users may need to run the install command with an extra flag:

python setup.py install --user

Docs

The speechmatics python SDK and CLI is documented at https://speechmatics.github.io/speechmatics-python/.

The Speechmatics API and product documentation can be found at https://docs.speechmatics.com.

Real-Time Client Usage

from speechmatics.models import *
import speechmatics

# Change to your own file
PATH_TO_FILE = "tests/data/ch.wav"
LANGUAGE = "en"

# Generate an API key at https://portal.speechmatics.com/manage-access/
API_KEY = ""

# Create a transcription client from config defaults
sm_client = speechmatics.client.WebsocketClient(API_KEY)

sm_client.add_event_handler(
    event_name=ServerMessageType.AddPartialTranscript,
    event_handler=print,
)

sm_client.add_event_handler(
    event_name=ServerMessageType.AddTranscript,
    event_handler=print,
)

conf = TranscriptionConfig(
    language=LANGUAGE, enable_partials=True, max_delay=5, enable_entities=True,
)

print("Starting transcription (type Ctrl-C to stop):")
with open(PATH_TO_FILE, "rb") as fd:
    try:
        sm_client.run_synchronously(fd, conf)
    except KeyboardInterrupt:
        print("\nTranscription stopped.")

Batch Client Usage

from speechmatics.models import ConnectionSettings, BatchTranscriptionConfig
from speechmatics.batch_client import BatchClient
from httpx import HTTPStatusError

API_KEY = "YOUR_API_KEY"
PATH_TO_FILE = "example.wav"
LANGUAGE = "en"

# Open the client using a context manager
with BatchClient(API_KEY) as client:
    try:
        job_id = client.submit_job(PATH_TO_FILE, BatchTranscriptionConfig(LANGUAGE))
        print(f'job {job_id} submitted successfully, waiting for transcript')

        # Note that in production, you should set up notifications instead of polling.
        # Notifications are described here: https://docs.speechmatics.com/features-other/notifications
        transcript = client.wait_for_completion(job_id, transcription_format='txt')
        # To see the full output, try setting transcription_format='json-v2'.
        print(transcript)
    except HTTPStatusError as e:
        if e.response.status_code == 401:
            print('Invalid API key - Check your API_KEY at the top of the code!')
        elif e.response.status_code == 400:
            print(e.response.json()['detail'])
        else:
            raise e

Example command-line usage

A complete list of commands and flags can be found in the SDK docs at https://speechmatics.github.io/speechmatics-python/.

Configuring Auth Tokens

• Setting an auth token for CLI authentication:

  speechmatics config set --auth-token $AUTH_TOKEN

  Auth tokens are stored in toml config at HOME_DIR/.speechmatics/config. You may also set the auth_token for each CLI command using the --auth-token flag. The --auth-token flag overrides the value stored in the config file, e.g.

  speechmatics transcribe --auth-token $AUTH_TOKEN example_audio.wav

• Removing an auth_token from the toml file:

  speechmatics config unset --auth-token

• Setting URLs for connecting to transcribers. These values can be used in places of the --url flag:

  speechmatics config set --rt-url wss://eu2.rt.speechmatics.com/v2 --batch-url https://asr.api.speechmatics.com/v2

• Unsetting transcriber URLs in the toml config:

  speechmatics config unset --rt-url --batch-url

• Setting URLs for connecting to transcribers. These values can be used in places of the --url flag:

  speechmatics config set --rt-url wss://eu2.rt.speechmatics.com/v2 --batch-url https://asr.api.speechmatics.com/v2

• Unsetting transcriber URLs in the toml config:

  speechmatics config unset --rt-url --batch-url

  Realtime ASR

• Starting a real-time session for self-service SaaS customers using a .wav file as the input audio:

  speechmatics transcribe --lang en example_audio.wav

• Real-time transcription of online stream (needs ffmpeg installed):

  ffmpeg -v 0 -i https://cdn.bitmovin.com/content/assets/art-of-motion-dash-hls-progressive/mpds/f08e80da-bf1d-4e3d-8899-f0f6155f6efa.mpd \
  -f s16le -ar 44100 -ac 1 -acodec pcm_s16le - | \
  speechmatics transcribe --raw pcm_s16le --sample-rate 44100 -
  

• Starting a real-time session for enterprise SaaS customers using a .wav file as the input audio:

  # Point URL to the a SaaS enterprise runtime
  URL=wss://neu.rt.speechmatics.com/v2/en
  
  speechmatics transcribe --url $URL example_audio.wav

• Starting a real-time session for on-prem customers using a .wav file as the input audio:

  # Point URL to the local instance of the realtime appliance
  URL=ws://realtimeappliance.yourcompany:9000/v2
  
  speechmatics transcribe --url $URL --lang en --ssl-mode none example_audio.wav

• Show the messages that are going over the websocket connection using verbose output:

  speechmatics -v transcribe --url $URL --ssl-mode none example_audio.wav

• The CLI also accepts an audio stream on standard input. Transcribe the piped input audio:

  cat example_audio.wav | speechmatics transcribe --url $URL --ssl-mode none -

• Pipe audio directly from the microphone (example uses MacOS with ffmpeg)

  List available input devices:

  ffmpeg -f avfoundation -list_devices true -i ""

  There needs to be at least one available microphone attached to your computer. The command below gets the microphone input and pipes it to the transcriber. You may need to change the sample rate to match the sample rate that your machine records at. You may also need to replace :default with something like :0 or :1 if you want to use a specific microphone.

  ffmpeg -loglevel quiet -f avfoundation -i ":default" -f f32le -acodec pcm_f32le -ar 44100 - \
  >   | speechmatics transcribe --url $URL --ssl-mode none --raw pcm_f32le --sample-rate 44100 -

• Transcribe in real-time with partials (example uses Ubuntu with ALSA). In this mode, the transcription engine produces words instantly, which may get updated as additional context becomes available.

  List available input devices:

  cat /proc/asound/cards

  Record microphone audio and pipe to transcriber:

  ffmpeg -loglevel quiet -f alsa -i hw:0 -f f32le -acodec pcm_f32le -ar 44100 - \
      | speechmatics transcribe --url $URL --ssl-mode none --enable-partials --raw pcm_f32le --sample-rate 44100 -

  Add the --print-json argument to see the raw JSON transcript messages being sent rather than just the plaintext transcript.

  Batch ASR

• Submit a .wav file for batch ASR processing

  speechmatics batch transcribe --lang en example_audio.wav

• Files may be submitted for asynchronous processing

  speechmatics batch submit example_audio.wav

• Enterprise SaaS and on-prem customers can point to a custom runtime:

  # Point URL to a custom runtime (in this case, the trial runtime)
  URL=https://trial.asr.api.speechmatics.com/v2/
  
  speechmatics batch transcribe --url $URL example_audio.wav

• Check processing status of a job

  # $JOB_ID is from the submit command output
  speechmatics batch job-status --job-id $JOB_ID

• Retrieve completed transcription

  # $JOB_ID is from the submit command output
  speechmatics batch get-results --job-id $JOB_ID

• Submit a job with automatic language identification

  speechmatics batch transcribe --language auto --langid-langs en,es example_audio.wav

  If Speechmatics is not able to identify a language with high enough confidence, the job will be rejected. This is to reduce the risk of transcribing incorrectly.

  --langid-langs is optional and specifies what language(s) you expect to be detected in the source files.

• Submit a job with translation

  speechmatics batch transcribe --translation-langs de,es --output-format json-v2 example_audio.wav

  --translation-langs is supported in asynchronous mode as well, and translation output can be retrieved using get-results with --output-format json-v2 set.

  When combining language identification with translation, we can't know if the identified language can be translated to your translation targets. If the translation pair is not supported, the error will be recorded in the metadata of the transcript.

• Start a real-time transcription with translation session using microphone audio and pipe to transcriber

  ffmpeg -loglevel quiet -f alsa -i hw:0 -f f32le -acodec pcm_f32le -ar 44100 - \
      | speechmatics rt transcribe --url $URL --ssl-mode none --raw pcm_f32le --sample-rate 44100 \
  --print-json --translation-langs fr -

• Submit a job with summarization

  speechmatics batch transcribe --summarize --output-format json-v2 example_audio.wav

• Submit a job with sentiment analysis

  speechmatics batch transcribe --sentiment-analysis --output-format json-v2 example_audio.wav

• Submit a job with topic detection

  speechmatics batch transcribe --detect-topics --output-format json-v2 example_audio.wav

• Submit a job with auto chapters

  speechmatics batch transcribe --detect-chapters --output-format json-v2 example_audio.wav

  Custom Transcription Config File

• Instead of passing all the transcription options via the command line you can also pass a transcription config file. The config file is a JSON file that contains the transcription options. The config file can be passed to the CLI using the --config-file option.

  speechmatics transcribe --config-file transcription_config.json example_audio.wav

• The format of this JSON file is described in detail in the Batch API documentation and RT API documentation.

SM Metrics

This package includes tooling for benchmarking transcription and diarization accuracy.

For more information, see the asr_metrics/README.md

Testing

To install development dependencies and run tests

pip install -r requirements-dev.txt
make test

Support

If you have any issues with this library or encounter any bugs then please get in touch with us at [email protected].

---

License: MIT