Get started with speech recognition using REST API

With the cloud-based Microsoft Speech Service, you can develop applications using REST API to convert spoken audio to text.

To use speech recognition REST API, the steps are as follows:

  1. Get a subscription key for using Speech API.
  2. Set the proper request header and send the request to the appropriate REST end points.
  3. Parse the response to get your transcribed text.


Subscribe to Speech API and get a free trial subscription key


You must have a subscription key before accessing Speech REST API.

Microsoft Speech API is part of Microsoft Cognitive Services on Azure(previously Project Oxford). You can get free trial subscription keys from the Cognitive Services Subscription page. After you select the Speech API, click Get API Key to get the key. It returns a primary and secondary key. Both keys are tied to the same quota, so you may use either key.

Precorded audio file

In this example, we use a recorded audio file to illustrate the usage of the REST API. Record a short audio file of you saying something short (for example: "What is the weather like today?" or "Find funny movies to watch."). The Microsoft speech recognition API also supports external microphone input.


The example requires that audio is recorded as wav file with PCM single channel (mono), 16 KHz.

Build recognition request and send it to the speech service

The next step for speech recognition is to send a POST request to the Microsoft Speech HTTP end points with proper request header and body.

HTTP endpoints

The URI of the HTTP endpoints for REST API is defined based on recognition mode and recognition language:<RECOGNITION_MODE>/cognitiveservices/v1?language=<LANGUAGE_TAG>&format=<OUTPUT_FORMAT>

<RECOGNITION_MODE> specifies the recognition mode, and must be of the following values: interactive, conversation, or dictation. It is a required resource path in the URI. For more information, see recognition modes.

<LANGUAGE_TAG> is a required parameter in the query string. It defines the target language for audio conversion. For example, en-US for English (United States). For more information, see recognition languages.

<OUTPUT_FOMAT> is an optional parameter in the query string. Its allowed values are simple and detailed. By default the service returns results in simple format. See output format for details.

Some examples of service URI are as follows.

Recognition mode Language Output format REST end point
interactive pt-BR default
conversation en-US detailed
dictation fr-FR simple


The HTTP service endpoints are only needed when your application uses REST APIs to call the speech service. If you use one of the client libraries, you usually do not need to know which endpoint is being used. The client libraries might use different service URIs, which are only applicable for the specific client library. For more information, see the client library of your choice.

Request headers

The follow fields must be set in the request header.

  • Ocp-Apim-Subscription-Key: Each time that you call the service, you must pass your subscription key in the Ocp-Apim-Subscription-Key header. The Microsoft Speech Service also supports passing authorization token instead of subscription key. See the Authentication page for details.
  • Content-type: The Content-type field describes the format and codec of the audio stream. Currently only wav file and PCM Mono 16000 encoding is supported, and the Content-type value for this format is audio/wav; codec=audio/pcm; samplerate=16000.

The field Transfer-Encoding is optional. Setting this field to chunked allows you to chop the audio into small chunks. For more information, see the page Chunked Transfer.

The follow is a sample request header.

Accept: application/json;text/xml
Content-Type: audio/wav; codec=audio/pcm; samplerate=16000
Ocp-Apim-Subscription-Key: YOUR_SUBSCRIPTION_KEY
Transfer-Encoding: chunked
Expect: 100-continue

Send request to the service

The following example shows how to send a speech recognition request to Microsoft speech REST end points. It uses interactive recognition mode.


Replace YOUR_AUDIO_FILE with the path to your prerecorded audio file, and YOUR_SUBSCRIPTION_KEY with your own subscription key.

$SpeechServiceURI =

# $OAuthToken is the authrization token returned by the token service.
$RecoRequestHeader = @{
  'Ocp-Apim-Subscription-Key' = 'YOUR_SUBSCRIPTION_KEY';
  'Transfer-Encoding' = 'chunked'
  'Content-type' = 'audio/wav; codec=audio/pcm; samplerate=16000'

# Read audio into byte array
$audioBytes = [System.IO.File]::ReadAllBytes("YOUR_AUDIO_FILE")

$RecoResponse = Invoke-RestMethod -Method POST -Uri $SpeechServiceURI -Headers $RecoRequestHeader -Body $audioBytes

# Show the result

Process speech recognition response

After processing the request, Microsoft Speech Service returns the results in a response as JSON format.


If the code above returns error, see Troubleshooting to locate the possible cause.

The following code snippet shows an example of how you can read the response from the stream.

# show the response in JSON format
ConvertTo-Json $RecoResponse

The following is a sample JSON response:

  "RecognitionStatus": "Success",
  "Offset": 22500000,
  "Duration": 21000000,
  "NBest": [{
    "Confidence": 0.941552162,
    "Lexical": "find a funny movie to watch",
    "ITN": "find a funny movie to watch",
    "MaskedITN": "find a funny movie to watch",
    "Display": "Find a funny movie to watch."


The Speech REST API has some limitations:

  • It only supports audio stream up to 15 seconds.
  • REST API does not support intermediate results during recognition. Users receive only the final recognition result.

To remove these limitations, you can use Microsoft speech client libraries or directly work with Speech WebSocket Protocol.

What's next


All Microsoft Cognitive Services SDKs and samples are licensed with the MIT License. For more information, see LICENSE.