rt/v1.1.0

What's Changed

• RT support for the next gen model by @mnemitz in #111

batch/v0.5.0

What's Changed

• Add model field to TranscriptionConfig for batch and RT by @mnemitz in #111
• Batch support for the next gen model by @rakeshv247 in #106
• Export new Model in batch sdk by @dumitrugutu in #114

New Contributors

• @mnemitz made their first contribution in #111
• @rakeshv247 made their first contribution in #106

batch/v0.4.9

What's Changed

• Fix JobError model: align with API spec, remove FetchDataError duplicate by @dumitrugutu in #109

batch/v0.4.10

What's Changed

• Add missing fields to RecognitionResult by @dumitrugutu in #110

batch/v0.4.8

What's Changed

• Batch SDK now exposes Speaker ID as part of transcript results. This means for Speaker ID and enrolment, you no longer have to resort to raw WebSocket plumbing, and can instead call the transcription results' .speakers property directly.

result = await client.transcribe(audio_file, transcription_config=config)
speakers = result.speakers

• Batch SDK now supports audio filtering config:

from speechmatics.batch import, TranscriptionConfig, AudioFilteringConfig                                                                                          
                                                                                                                                                                               
  config = TranscriptionConfig(                                                                                                                                                  
      language="en",                                                                                                                                                             
      audio_filtering_config=AudioFilteringConfig(volume_threshold=10),                                                                                                          
  )  

New Contributors

• @J-Jaywalker made their first contribution in #105

Full Changelog: batch/v0.4.7...batch/v0.4.8

rt/v1.0.0

Breaking Change

Add an extra optional attribute** to the ForceEndOfUtterance message to help the engine know the real-world time marker for when the forced end of utterance should be considered. It should only be used with transcribers that support this extended message.

{"message": "ForceEndOfUtterance", "timestamp": 1.23}

Using the client.audio_seconds_sent property will return the number of seconds of audio that have been sent to the transcriber, calculated using encoding format and sample rate of the session's audio (AudioFormat). This can then be used when calling the force end of utterance:

client.force_end_of_utterance(timestamp=client.audio_seconds_sent)

Transcribers that don't support the new timestamp parameter will return an error:

Unable to validate message against the API JSON schema: ["<ValidationError: '.: {'message': 'ForceEndOfUtterance', 'timestamp': 1.87} is not valid under any of the given schemas'>"]

It can be disabled by passing in the function timestamp=None as such:

client.force_end_of_utterance(timestamp=None)

• Added timestamp attribute to ForceEndOfUtterance by @sam-s10s in #84

New Contributors

• @giorgosHadji made their first contribution in #95

batch/v0.4.7

What's Changed

• add support for requesting parallel engines in a http batch job by @giorgosHadji in #95
• add user_id as input to header for http batch V2 API by @giorgosHadji in #97

Usage with:

from speechmatics.batch import AsyncClient

async def main():
    async with AsyncClient(url='http://localhost:PORT/v2') as client:
        result = await client.transcribe(audio_file_path, parallel_engines=2, user_id="MY_USER_ID")
        print(result)

asyncio.run(main())

which will flow into the headers being sent as part of the POST request to the server.

These features are available for use for on Prem scenarios only, given the transcriber version supports them. If it doesn't support them, they will be ignored.

The need for user_id is specific for the speaker_id feature as its a requirement to be able to use it - https://docs.speechmatics.com/speech-to-text/features/speaker-identification

New Contributors

• @giorgosHadji made their first contribution in #95

batch/v0.4.6

What's Changed

• Add transcript filtering config and fix output config deserialization by @lgavincrl in #88
• Fix import TranscriptFilteringConfig in batch sdk by @dumitrugutu in #91

batch/v0.4.5

What's Changed

• Add OutputConfig option in JobConfig for batch sdk by @dumitrugutu in #80

voice/v0.2.8

What's Changed

• Update Scribe Preset Configuration by @LArmstrongDev in #77
• Fix for short utterances when using ForceEndOfUtterance by @sam-s10s in #78