Skip to content

Latest commit

 

History

History
110 lines (71 loc) · 4.18 KB

challenge_03.md

File metadata and controls

110 lines (71 loc) · 4.18 KB
Raw

Hints for Challenge 3

Create a new Python 3.6 Notebook in Azure Notebooks. Next, create a Speech API Key in the Azure Portal:

alt text

As region, we'll be using West Europe in this example. You can find your API key under the service, then Keys.

You can use this file test.wav for testing.

Text-to-Speech

First, we need to request a token from the Issue Token endpoint of the Speech API. Each token is valid for 10 minutes, hence we can either reuse it multiple times (to minimize network traffic and latency), or request a new one for each call:

import requests, json
import IPython.display as ipd

api_key = "xxxx" # Enter your API key here

token_url = "https://westeurope.api.cognitive.microsoft.com/sts/v1.0/issuetoken"
headers = {'Ocp-Apim-Subscription-Key': api_key}

response = requests.post(token_url, headers=headers)
token = response.text

print("Token: " + token)

Once we have the token, we can form our request for generating speech:

url = "https://westeurope.tts.speech.microsoft.com/cognitiveservices/v1"
headers = {'Authorization': token,
           'Content-Type': 'application/ssml+xml',
           'User-Agent': 'Test',
           'X-Microsoft-OutputFormat': 'riff-16khz-16bit-mono-pcm'}

data = "<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xml:lang='en-US'> \
<voice name='Microsoft Server Speech Text to Speech Voice (en-US, JessaRUS)'> \
    Hello, welcome to the Cognitive Services Bootcamp!  \
</voice></speak>"

response = requests.post(url, headers=headers, data=data)
audio_data = response.content

print(response.headers)

We can just write it out to a *.wav file and then download or play it:

with open("test.wav", "wb") as f: 
    f.write(audio_data)
    
ipd.Audio('test.wav')

There are many different voices available to choose from. By updating the XML request, we can easily specify a different voice or language. From here on, it should be easy to generate German speech.

Speech-to-Text

Let's take the generated or provided test.wav from the example before and convert it back to text. Again, let's first create a token:

import requests, json

api_key = "xxx" # Enter your API key here

token_url = "https://westeurope.api.cognitive.microsoft.com/sts/v1.0/issuetoken"
headers = {'Ocp-Apim-Subscription-Key': api_key}

response = requests.post(token_url, headers=headers)
token = response.text

print("Token: " + token)

Now that we have a token, we can call the speech-to-text endpoint and include the wav data:

url = "https://westeurope.stt.speech.microsoft.com/speech/recognition/conversation/cognitiveservices/v1"

headers = {'Authorization': 'Bearer ' + token,
           'Accept': 'application/json',
           'Ocp-Apim-Subscription-Key': api_key,
           'Content-Type': 'audio/wav; codec=audio/pcm; samplerate=16000'}

params = {'language': 'en-US', 'format': 'detailed'}

with open("test.wav", 'rb') as f:
    data = f.read()

response = requests.post(url, headers=headers, params=params, data=data)
print(json.dumps(response.json(), indent=2))

For recognizing longer text with multiple sentences, you can follow the following tutorial.

Note:

As of May 2019, also compressed audio is supported (e.g., MP3s), see here,

Besides that, the speech-to-text API expects audio with the following specifics:

  • 16-bit WAV format with PCM or OGG format with OPUS
  • Single channel (mono) at 8 or 16 KHz

More details, see here.

Now that we've converted the user's speech into text, we can detect the intent of the text in the next challenge!