Power Platform Community Forum Thread Details

Hi everyone,

We are currently building a Copilot Studio agent with voice enablement to support multilingual. After some research, I was able to build and host the agent with SSO and voice capabilities using Azure Speech Services.

At present, I’m embedding the chatbot into a custom website using plain JavaScript and hosting it on Azure Static Web Apps. As we move forward, I’m looking for additional documentation, resources, or videos that provide deeper insight and more control over the following areas.

Integration and configuration options when embedding a Copilot Studio agent into a custom web application

Best practices for voice implementation, especially when using Azure Speech Services

Guidance on choosing between Azure Static Web Apps and Azure Web Apps for this scenario, including differences in configuration flexibility, update strategies, security, and integration capabilities

Azure Speech Service configurations, such as fine-tuning voices, improving naturalness, making conversations more contextual and conversational, and handling multilingual scenarios effectively

I’ve already reviewed the available Microsoft documentation, but I’m still struggling to find detailed, end-to-end guidance that covers these aspects—this may simply be due to gaps in my understanding.

Any recommendations, real-world examples, or best practices from the community would be greatly appreciated.

Thank you in advance for your guidance!

Categories:

Copilot Studio skills development

Publish & channel management

General topics

Hello Srijapam,
1. Embedding an Agent in a Custom Web App

Leverage the Bot Framework Web Chat SDK with Direct Line to maintain full control. Always generate tokens server-side to keep secrets secure; never expose them on the client side.

Docs: Configure web and Direct Line channel security - Microsoft Copilot Studio | Microsoft Learn
2. Voice Integration Best Practices
For speech-to-text, use AutoDetectSourceLanguageConfig to support multiple languages. For text-to-speech, apply SSML tags like <prosody>, <break>, and <mstts:express-as> for more natural output. Multilingual neural voices, such as JennyMultilingualNeural, can handle over 100 languages.
Docs: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-synthesis-markup
3. Static Web Apps vs App Service
Choose Azure Web Apps or Static Web Apps with an Azure Functions backend to securely issue Speech and Direct Line tokens. Avoid pure client-side static hosting, as it risks exposing API keys.

4. Enhancing Voice Naturalness
Utilize SSML styles (e.g., friendly, empathetic) through <mstts:express-as>, add pauses with <break>, and adjust speech rate or pitch using <prosody>. For brand-specific voices, consider Custom Neural Voice.
Docs: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/custom-neural-voice