Speech recognition

There are two types of speech recognition available:

• recognition of audio files;
• recognition of streaming audio.

To recognize speech from an audio file, send the audio file in the body of the POST request to https://voice.mcs.mail.ru/asr, specifying the correct Content -Type in header.

Request example:

Copy
curl -L --request POST 'https://voice.mcs.mail.ru/asr'--header 'Content-Type: audio/ogg; codecs=opus'--header 'Authorization: Bearer xxxxxxxxxx'--data-binary '@/Users/User/tts.ogg'

Answer example:

Copy
{	"qid": "0ac6294a351d42ad859404ecd349e4b9",	"result": {		"texts": [			{				"text": "hello alice",				"confidence": 1.0,				"punctuated_text": "Hello, Alice."			}		],		"phrase_id": "20220921-1515-4d75-92b4-24b6c101ba6a"	}}

To recognize a chunk (small piece of speech), you need to send a request to create a task. After that, it will be possible to send chunks and receive the final result.

In order to create a task, it is enough to send a POST request to https://voice.mcs.mail.ru/asr_stream/create_task with an authorization header with access_token, the response will be task_id, task_token.

Request example:

Copy
curl --request POST \  --url https://voice.mcs.mail.ru/asr_stream/create_task  --header 'Authorization: Bearer access_tokenxxxxxxxx'

Answer example:

Copy
{  "qid": "61b5cf067c494b4a9a0b87a3c43e37ef",  "result": {    "task_id": "05ad987e-ceee-4186-acdb-956148b91692",    "task_token": "040b2fcfc3d9b9806b691070e873125dfc0450a8251887ba91b19be08eb3951c"  }}

A chunk is an audio fragment of the selected format, respectively, headers must be present in each chunk.

To send a chunk, all you need to do is:

• send a POST request to https://voice.mcs.mail.ru/asr_stream/add_chunk, passing in the Authorization-task_token header;
• pass task_id and chunk_num in GET parameters (numbering starts from 0);
• specify the correct Content-Type in the request header.
• a chunk is sent in the request body, which is an array of bytes in wav or ogg format.

The response will be the result of chunk recognition.

Request example:

Copy
curl --request POST \  --url 'https://voice.mcs.mail.ru/asr_stream/add_chunk?task_id=xxxxx&chunk_num=2' \  --header 'Authorization: Bearer task_tokenxxxxxxxx' \  --header 'Content-Type: audio/wave' \  --data 'xxxxxxxxxx'

Answer example:

Copy
{  "qid": "4d44cb0eb81f4e7f84a7997ec4f2f3c4",  "result": {    "text": "hello marusya"  }}

At any time after sending chunks, you can get the result, for this you need to send a GET request to https://voice.mcs.mail.ru/asr_stream/get_result, passing Authorization-task_token in the header, task_id in the GET parameters.

The response will include the recognition result with the current status of the task.

Request example:

Copy
curl --request GET \  --url 'https://voice.mcs.mail.ru/asr_stream/get_result?task_id=xxxxx' \  --header 'Authorization: Bearer task_tokenxxxxxxxx' \

Answer example:

Copy
{  "qid": "517e5ba9f4a9465c9d73778bedac0808",  "result": {    "text": "hello marusya hello marusya",    "status": "done"  }}