Ender Turing

Ender Turing is a solution for voice content understanding, analytics and business insights. Check enderturing.com for details.

Installation

$ pip install enderturing

For using streaming speech recognition functions, you'll also need FFmpeg installed.

Ubuntu:

$ sudo apt install ffmpeg

MacOS homebrew:

$ brew install ffmpeg

For other OS, please follow FFmpeg installation guides.

Quick Start

import asyncio
from enderturing import Config, EnderTuring, RecognitionResultFormat

# create configuration
config = Config.from_url("https://admin%40local.enderturing.com:your_password@enterturing.yourcompany.com")
et = EnderTuring(config)

# access sessions list
sessions = et.sessions.list()
print(sessions)

# get recognizer for one of configured languages
recognizer = et.get_speech_recognizer(language='en')

async def run_stream_recog(f, r, result_format):
    async with r.stream_recognize(f, result_format=result_format) as rec:
        text = await rec.read()
    return text

# recognize specified file
loop = asyncio.get_event_loop()
task = loop.create_task(run_stream_recog("my_audio.mp3", recognizer, result_format=RecognitionResultFormat.text))
loop.run_until_complete(task)
print(task.result())

Usage

SDK contains two major parts:

• Using Ender Turing REST API
• Speech recognition

Using Ender Turing API

All API calls are accessible via an instance or EnderTuring. API methods are grouped, and each group is a property of EnderTuring. Examples:

from enderturing import Config, EnderTuring, RecognitionResultFormat

et = EnderTuring(Config.from_env())

# access sessions list
sessions = et.sessions.list()

# working with ASR
et.asr.get_instances(active_only=True)

# accessing raw json
et.raw.create_event(caller_id='1234', event_data={"type": "hold"})

Access Configuration

To access API, you need to know an authentication key (login), authentication secret (password), and installation URL (e.g. https://enderturing.yourcompany.com/)

There are multiple ways to pass config options:

• from environmental variables (Config.from_env()).
• creating Config with parameters (e.g. Config(auth_key="my_login", auth_secret="my_secret""))
• using Enter Turing configuration URL (Config.from_url())

Creating Speech Recognizer

There two options to create a speech recognizer:

If you have access to API configured:

recognizer = et.get_speech_recognizer(language='en')

If you know URL and sample rate of desired ASR instance:

from enderturing import AsrConfig, SpeechRecognizer

config = AsrConfig(url="wss://enderturing", sample_rate=8000)
recognizer = SpeechRecognizer(config)

Recognizing a File

SpeechRecognizer.recognize_file method returns an async text stream. Depending on parameters, each line contains either a text of utterance or serialized JSON.

If you are only interested in results after recognition is complete, you can use the read() method. E.g.

async with recognizer.recognize_file("my_audio.wav", result_format=RecognitionResultFormat.text) as rec:
    text = await rec.read()

If you prefer getting words and phrases as soon as they are recognized - you can use the readline() method instead. E.g.

async with recognizer.recognize_file(src, result_format=RecognitionResultFormat.jsonl) as rec:
    line = await rec.readline()
    while line:
        # Now line contains a json string, you can save it or do something else with it
        line = await rec.readline()

Working With Multichannel Audio

If an audio file has more than one channel - by default system will recognize each channel and return a transcript for each channel. To change the default behavior - you can use channels parameter of SpeechRecognizer.recognize_file. Please check method documentation for details.

Sometimes an audio is stored as a file per channel, e.g., contact center call generates two files: one for a client and one for a support agent. But for analysis, it's preferable to see transcripts of the files merged as a dialog. In this scenario, you can use recognizer.recognize_joined_file([audio1, audio2]).

License

Released under the MIT license.