TTS Rejecting Certain Characters

Question

TTS Rejecting Certain Characters

SB-1234 10

I've noticed that words are rejected (error 400) like this example:

Курица - rejected because all of the letters are cyrillic (unicode)

Курицa - allowed because the 'a' is ascii

Is there an unwritten rule somewhere? How should I TTS phrases that are entirely non-ascii?

My headers etc are properly formed, I only spotted this when I noticed certain results failing

Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-07-27T11:58:56.7133333+00:00

@SB-1234 Are you using only text or SSML with TTS? I have tried the speech studio to check if the rejected word can be synthesized and it seems to work.
Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-07-31T13:49:09.68+00:00

@SB-1234 Did you get to try the same with the speech studio?
SB-1234 10 Reputation points

2023-09-02T20:16:41.89+00:00
@Rohit Mungi

Sorry I didn't get notified of your response and have been away on holiday

I use SSML, here is an example:

<speak version='1.0' xml:lang='ru-RU'> <voice xml:lang='ru-RU' xml:gender='Female' name='ru-RU-DariyaNeural'> Курица </voice> </speak>

This works fine in the speech studio, but it fails when submitted using Python's requests module. Other requests work fine if they contain more than 1 ASCII character as I mentioned in the OP

SB-1234 10

Failing request (sanitised):

Header:

{'User-Agent': 'XXXX', 'Accept-Encoding': 'gzip, deflate', 'Accept': '*/*', 'Connection': 'keep-alive', 'X-Microsoft-OutputFormat': 'riff-44100hz-16bit-mono-pcm', 'Content-Type': 'application/ssml+xml', 'Ocp-Apim-Subscription-Key': 'XXXX', 'Content-Length': '169'}

Body:


    <speak version='1.0' xml:lang='ru-RU'>
    <voice xml:lang='ru-RU' xml:gender='Female' name='ru-RU-DariyaNeural'>
            Медведь
    </voice>
    </speak>

Successful request:

Header:

{'User-Agent': 'XXXX', 'Accept-Encoding': 'gzip, deflate', 'Accept': '*/*', 'Connection': 'keep-alive', 'X-Microsoft-OutputFormat': 'riff-44100hz-16bit-mono-pcm', 'Content-Type': 'application/ssml+xml', 'Ocp-Apim-Subscription-Key': 'XXXX', 'Content-Length': '165'}

Body:


    <speak version='1.0' xml:lang='ru-RU'>
    <voice xml:lang='ru-RU' xml:gender='Female' name='ru-RU-DariyaNeural'>
            Кот
    </voice>
    </speak>

1 answer

Your answer

Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-07-27T11:58:56.7133333+00:00

@SB-1234 Are you using only text or SSML with TTS? I have tried the speech studio to check if the rejected word can be synthesized and it seems to work.
Rohit Mungi 49,131 Reputation points Microsoft Employee Moderator

2023-07-31T13:49:09.68+00:00

@SB-1234 Did you get to try the same with the speech studio?
SB-1234 10 Reputation points

2023-09-02T20:16:41.89+00:00

@Rohit Mungi

Sorry I didn't get notified of your response and have been away on holiday

I use SSML, here is an example:

<speak version='1.0' xml:lang='ru-RU'> <voice xml:lang='ru-RU' xml:gender='Female' name='ru-RU-DariyaNeural'> Курица </voice> </speak>

This works fine in the speech studio, but it fails when submitted using Python's requests module. Other requests work fine if they contain more than 1 ASCII character as I mentioned in the OP
SB-1234 10 Reputation points

2023-09-02T20:51:03.1833333+00:00

Failing request (sanitised):

Header:

{'User-Agent': 'XXXX', 'Accept-Encoding': 'gzip, deflate', 'Accept': '*/*', 'Connection': 'keep-alive', 'X-Microsoft-OutputFormat': 'riff-44100hz-16bit-mono-pcm', 'Content-Type': 'application/ssml+xml', 'Ocp-Apim-Subscription-Key': 'XXXX', 'Content-Length': '169'}

Body:

<speak version='1.0' xml:lang='ru-RU'> <voice xml:lang='ru-RU' xml:gender='Female' name='ru-RU-DariyaNeural'> Медведь </voice> </speak>

Successful request:

Header:

{'User-Agent': 'XXXX', 'Accept-Encoding': 'gzip, deflate', 'Accept': '*/*', 'Connection': 'keep-alive', 'X-Microsoft-OutputFormat': 'riff-44100hz-16bit-mono-pcm', 'Content-Type': 'application/ssml+xml', 'Ocp-Apim-Subscription-Key': 'XXXX', 'Content-Length': '165'}

Body:

<speak version='1.0' xml:lang='ru-RU'> <voice xml:lang='ru-RU' xml:gender='Female' name='ru-RU-DariyaNeural'> Кот </voice> </speak>

Answer 1

SB-1234 10

I've fixed the problem, it's actually a bug in Python's requests module. The content-length was miscalculated, and it seems to be optional for TTS

I remove the content-length like so, and now everything works:

    session = requests.Session()
    request = requests.Request('POST', url=api_endpoint, headers=headers, data=data)
    prepped_request = request.prepare()

    del prepped_request.headers['Content-Length']

    response = session.send(prepped_request)

0 comments

Share via

TTS Rejecting Certain Characters

1 answer

Your answer