Undesired Pause in Neural Voice Synthesis for Long Sentences

Question

Undesired Pause in Neural Voice Synthesis for Long Sentences

佳鑫朱 0

Same as this https://learn.microsoft.com/en-us/answers/questions/edit/1331368/answer/1391759?orderby=helpful#answer-1391759

I pronounce people in Chinese“云希”，I had the same problem. When I read that sentence,"没想到小三竟自导自演了一场戏" I always pause when I get to the third word.

My request was to read the novel in a tight manner, without requiring too many emotional and anthropomorphic pauses. Is there any solution that can meet my requirements?

I've been through the documentation, there's no compact mode, I don't need autonomous pauses, I don't need anthropomorphism. Like no pause before the full stop. What should I do?

dupammi 8,615 Microsoft External Staff

Hi @佳鑫朱 ,

Thank you for using the Microsoft Q&A.

To remove the pauses before periods or full stops in Chinese, you can use the break element with the strength attribute set to "none" or "time" attribute set to "1ms". This will indicate that no pause should be inserted before the full stop.

Here is a sample that I tried using the break element, to remove pause before full stop for English language.

import azure.cognitiveservices.speech as speechsdk

# Set the subscription key and region for the custom model
speech_key = "027cd6b3edf942449266254b327de65e"
service_region = "westeurope"
# Set the endpoint ID for the custom model
endpoint_id = "b0fb8f59-8419-440d-9e14-8c63d1035b21"
# Create a SpeechConfig object with the subscription key and region
speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)
speech_config.endpoint_id = endpoint_id
# Create a SpeechSynthesizer object with the SpeechConfig object
speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=speech_config)
# Create a SpeechSynthesizer object
speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)
#speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=speech_config)
speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=speech_config)
ssml = """<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xml:lang="en-US">
    <voice name="zh-CN-XiaochenNeural">
    <prosody rate="medium">
      <p>
        This is a sentence without a pause before the full stop<break strength="none"/>This is another
      </p>
    </prosody>
  </voice>
</speak>
"""
# prosody rate can be fast, medium or slow, as per your project #requirements.
result = speech_synthesizer.speak_ssml_async(ssml).get()

I was not able to test Chinese sentence, due to language barrier.

Please try at your end if it meets your project requirement.

I hope this helps. Thank you!

dupammi 8,615 Reputation points Microsoft External Staff

2023-12-16T00:40:28.45+00:00

Hi @佳鑫朱 ,

We haven’t heard from you on the last response and was just checking back to see if you got chance to check my above response.

Thank You!
佳鑫朱 0 Reputation points

2023-12-16T05:17:57.7533333+00:00

@dupammi

Hello, you don’t seem to understand what I mean. I mean read the entire article quickly, without any pauses except periods. I don't do the break manually. If you look at the first line of my question, others also have this need. It is impossible for me to add <break strength="none"/> in the middle of each word. I want to read it in compact mode at once, without the need for self-service pauses.
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-16T07:32:44.29+00:00

Hi @佳鑫朱 ,

Thank you for clarifying.

Unfortunately, there is no way to achieve the requirement of reading the entire article quickly without any pauses except for periods using the Text-to-Speech service in Azure.

I hope this information helps.
佳鑫朱 0 Reputation points

2023-12-16T16:27:10.3466667+00:00

@dupammi Hello, I don't understand what you mean, does it mean that there is no way to fulfill my needs? I'm using text-to-speech
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-17T00:20:13.8833333+00:00

Hi @佳鑫朱 ,

Apologies for the inconvenience you experienced.

Thank you for your feedback and suggestion to add a feature to achieve the requirement of reading the entire Novel/Article quickly without any pauses, except for periods (Full stops) using the Text-to-Speech service in Azure. Unfortunately, this feature is not currently available in TTS. However, we appreciate your feedback and encourage you to submit a feature request for the same through the Azure Feedback Forum. Here's the link to the Azure Feedback Forum: Post idea · Community (azure.com)

I hope you understand. Do let us know if you have any further queries.

Your answer

dupammi 8,615 Reputation points Microsoft External Staff

2023-12-15T12:19:07.73+00:00

Hi @佳鑫朱 ,

Thank you for using the Microsoft Q&A.

To remove the pauses before periods or full stops in Chinese, you can use the break element with the strength attribute set to "none" or "time" attribute set to "1ms". This will indicate that no pause should be inserted before the full stop.

Here is a sample that I tried using the break element, to remove pause before full stop for English language.

import azure.cognitiveservices.speech as speechsdk # Set the subscription key and region for the custom model speech_key = "027cd6b3edf942449266254b327de65e" service_region = "westeurope" # Set the endpoint ID for the custom model endpoint_id = "b0fb8f59-8419-440d-9e14-8c63d1035b21" # Create a SpeechConfig object with the subscription key and region speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region) speech_config.endpoint_id = endpoint_id # Create a SpeechSynthesizer object with the SpeechConfig object speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=speech_config) # Create a SpeechSynthesizer object speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region) #speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=speech_config) speech_synthesizer = speechsdk.SpeechSynthesizer(speech_config=speech_config) ssml = """<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xml:lang="en-US"> <voice name="zh-CN-XiaochenNeural"> <prosody rate="medium"> <p> This is a sentence without a pause before the full stop<break strength="none"/>This is another </p> </prosody> </voice> </speak> """ # prosody rate can be fast, medium or slow, as per your project #requirements. result = speech_synthesizer.speak_ssml_async(ssml).get()

I was not able to test Chinese sentence, due to language barrier.

Please try at your end if it meets your project requirement.

I hope this helps. Thank you!
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-16T00:40:28.45+00:00

Hi @佳鑫朱 ,

We haven’t heard from you on the last response and was just checking back to see if you got chance to check my above response.

Thank You!
佳鑫朱 0 Reputation points

2023-12-16T05:17:57.7533333+00:00

@dupammi

Hello, you don’t seem to understand what I mean. I mean read the entire article quickly, without any pauses except periods. I don't do the break manually. If you look at the first line of my question, others also have this need. It is impossible for me to add <break strength="none"/> in the middle of each word. I want to read it in compact mode at once, without the need for self-service pauses.
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-16T07:32:44.29+00:00

Hi @佳鑫朱 ,

Thank you for clarifying.

Unfortunately, there is no way to achieve the requirement of reading the entire article quickly without any pauses except for periods using the Text-to-Speech service in Azure.

I hope this information helps.
佳鑫朱 0 Reputation points

2023-12-16T16:27:10.3466667+00:00

@dupammi Hello, I don't understand what you mean, does it mean that there is no way to fulfill my needs? I'm using text-to-speech
dupammi 8,615 Reputation points Microsoft External Staff

2023-12-17T00:20:13.8833333+00:00

Hi @佳鑫朱 ,

Apologies for the inconvenience you experienced.

Thank you for your feedback and suggestion to add a feature to achieve the requirement of reading the entire Novel/Article quickly without any pauses, except for periods (Full stops) using the Text-to-Speech service in Azure. Unfortunately, this feature is not currently available in TTS. However, we appreciate your feedback and encourage you to submit a feature request for the same through the Azure Feedback Forum. Here's the link to the Azure Feedback Forum: Post idea · Community (azure.com)

I hope you understand. Do let us know if you have any further queries.

Share via

Undesired Pause in Neural Voice Synthesis for Long Sentences

Your answer