Use personal voice in your application

Raksts
09/20/2024

You can use the speaker profile ID for your personal voice to synthesize speech in any of the 91 languages supported across 100+ locales. A locale tag isn't required. Personal voice uses automatic language detection at the sentence level.

Integrate personal voice in your application

You need to use speech synthesis markup language (SSML) to use personal voice in your application. SSML is an XML-based markup language that provides a standard way to mark up text for the generation of synthetic speech. SSML tags are used to control the pronunciation, volume, pitch, rate, and other attributes of the speech synthesis output.

The speakerProfileId property in SSML is used to specify the speaker profile ID for the personal voice.
The voice name is specified in the name property in SSML. For personal voice, the voice name must be one of the supported base model voice names. To get a list of supported base model voice names, use the BaseModels_List operation of the custom voice API.

Note

The voice names labeled with the Latest, such as DragonLatestNeural or PhoenixLatestNeural, will be updated from time to time; its performance may vary with updates for ongoing improvements. If you would like to use a fixed version, select one labeled with a version number, such as PhoenixV2Neural.
DragonLatestNeural is a base model with superior voice cloning similarity compared to PhoenixLatestNeural. PhoenixLatestNeural is a base model with more accurate pronunciation and lower latency than DragonLatestNeural.
For personal voice, you can use the <lang xml:lang> element to adjust the speaking language. It's the same as with multilingual voices. See how to use the lang element to speak different languages.

Here's example SSML in a request for text to speech with the voice name and the speaker profile ID. The sample also demonstrates how to switch languages from en-US to zh-HK using the <lang xml:lang> element.

<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xmlns:mstts='http://www.w3.org/2001/mstts' xml:lang='en-US'>
    <voice name='DragonLatestNeural'> 
        <mstts:ttsembedding speakerProfileId='your speaker profile ID here'> 
            I'm happy to hear that you find me amazing and that I have made your trip planning easier and more fun. 
            <lang xml:lang='zh-HK'>我很高興聽到你覺得我很了不起，我讓你的旅行計劃更輕鬆、更有趣。</lang>
        </mstts:ttsembedding> 
    </voice> 
</speak>

You can use the SSML via the Speech SDK or REST API.

Real-time speech synthesis: Use the Speech SDK or REST API to convert text to speech.
- When you use Speech SDK, don't set Endpoint ID, just like prebuild voice.
- When you use REST API, please use prebuilt neural voices endpoint.

Supported and unsupported SSML elements for personal voice

For detailed information on the supported and unsupported SSML elements for Phoenix and Dragon models, refer to the following table. For instructions on how to use SSML elements, refer to the SSML document structure and events.

Element	Description	Supported in Phoenix	Supported in Dragon
`<voice>`	Specifies the voice and optional effects (`eq_car` and `eq_telecomhp8k`).	Yes	Yes
`<mstts:express-as>`	Specifies speaking styles and roles.	No	No
`<mstts:ttsembedding>`	Specifies the `speakerProfileId` property for a personal voice.	Yes	Yes
`<lang xml:lang>`	Specifies the speaking language.	Yes	Yes
`<prosody>`	Adjusts pitch, contour, range, rate, and volume.
`pitch`	Indicates the baseline pitch for the text.	No	No
`contour`	Represents changes in pitch.	No	No
`range`	Represents the range of pitch for the text.	No	No
`rate`	Indicates the speaking rate of the text.	Yes	Yes
`volume`	Indicates the volume level of the speaking voice.	No	No
`<emphasis>`	Adds or removes word-level stress for the text.	No	No
`<audio>`	Embeds prerecorded audio into an SSML document.	Yes	No
`<mstts:audioduration>`	Specifies the duration of the output audio.	No	No
`<mstts:backgroundaudio>`	Adds background audio to your SSML documents or mixes an audio file with text to speech.	Yes	No
`<phoneme>`	Specifies phonetic pronunciation in SSML documents.
`ipa`	One of the phonetic alphabets.	Yes	No
`sapi`	One of the phonetic alphabets.	No	No
`ups`	One of the phonetic alphabets.	Yes	No
`x-sampa`	One of the phonetic alphabets.	Yes	No
`<lexicon>`	Defines how multiple entities are read in SSML.	Yes	Yes (only support alias)
`<say-as>`	Indicates the content type, such as number or date, of the element's text.	Yes	Yes
`<sub>`	Indicates that the alias attribute's text value should be pronounced instead of the element's enclosed text.	Yes	Yes
`<math>`	Uses the MathML as input text to properly pronounce mathematical notations in the output audio.	Yes	No
`<bookmark>`	Gets the offset of each marker in the audio stream.	Yes	No
`<break>`	Overrides the default behavior of breaks or pauses between words.	Yes	Yes
`<mstts:silence>`	Inserts pauses before or after text, or between two adjacent sentences.	Yes	No
`<mstts:viseme>`	Defines the position of the face and mouth while a person is speaking.	Yes	No
`<p>`	Denotes paragraphs in SSML documents.	Yes	Yes
`<s>`	Denotes sentences in SSML documents.	Yes	Yes

Supported and unsupported SDK features for personal voice

The following table outlines which SDK features are supported for Phoenix and Dragon models. For details on how to utilize these SDK features in your applications, refer to Subscribe to synthesizer events.

SDK features	Description	Supported in Phoenix	Supported in Dragon
Word boundary	Signals that a word boundary was received during synthesis, providing precise word timing during the speech synthesis process.	Yes	No
Viseme events	Provides viseme (lips, jaw, and tongue movement) information during synthesis, allowing visual synchronization.	Yes	No

Reference documentation

Custom voice REST API reference documentation

Next steps

Learn more about custom neural voice in the overview.
Learn more about Speech Studio in the overview.

Kopīgot, izmantojot