Skip to content

Azure AI Speech Translation Guide

🎯 Problem

You need to implement a real-time speech-to-text translation solution, possibly with speech output in the target language(s).

✅ Solution with Azure

Use Azure AI Speech Translation service via the Speech SDK, which enables real-time speech recognition, translation, and synthesis into one or more target languages.

🧩 Componenti richiesti

  • ✅ Azure AI Speech resource (or Azure AI Services resource)
  • ✅ Azure AI Speech SDK (supports C#, Python, etc.)
  • ✅ SpeechTranslationConfig object
  • ✅ TranslationRecognizer object
  • ✅ Optional: SpeechSynthesizer (for audio output)
  • ✅ Optional: AudioConfig (for specifying input/output source)

🛠️ Architettura / Sviluppo

🔹 Provision Azure Speech Resource

  • Create a Speech resource in Azure
  • Retrieve:
  • Location (e.g., eastus)
  • Key (from Keys and Endpoint section in Azure Portal)

🔹 Speech-to-Text Translation

Flow using SDK: 1. Create SpeechTranslationConfig using your Azure key + region 2. Specify: * Source (speech recognition) language * One or more target languages 3. (Optional) Use AudioConfig for custom audio input 4. Create a TranslationRecognizer with the above 5. Call RecognizeOnceAsync() to: * Transcribe spoken input * Translate it into target language(s) 6. Use result properties: * Text: original transcription * Translations: dictionary of target language translations * Reason: check if translation is successful (e.g., RecognizedSpeech)

🔹 Synthesize Translations (Speech-to-Speech)

🟣 Event-Based Synthesis

  • Use only for 1:1 translation
  • In TranslationConfig, set the voice for target speech
  • Add handler for Synthesizing event on TranslationRecognizer
  • In handler: call .GetAudio() to get translated audio stream

🟢 Manual Synthesis

  • Use for 1:n translations (multiple targets)
  • Steps:
  • Use TranslationRecognizer → get Translations dictionary
  • For each translation:
    • Create a SpeechSynthesizer
    • Call SpeakTextAsync() for each translated string

🧠 Best Practice / Considerazioni

  • 🔐 Always secure and store resource keys and endpoints properly
  • 🌐 Use standard ISO language codes (e.g., en, fr)
  • 🧪 Test voice compatibility when using neural voices in synthesis
  • ⚡ Event-based synthesis better for low-latency 1:1 flows
  • 🔄 Use manual synthesis for multi-language or post-processed translations

❓ Domande simulate d'esame

1.

Q: What configuration object is required to define the source and target languages for speech translation in Azure AI Speech SDK?
A: SpeechTranslationConfig

2.

Q: What object in the Azure Speech SDK is used to retrieve translated audio via an event handler?
A: TranslationRecognizer with a Synthesizing event handler

3.

Q: Which result property contains the translated text for each target language?
A: Translations (dictionary in the result object)

4.

Q: How can you perform speech-to-speech translation for multiple target languages without handling events?
A: Use manual synthesis: iterate through Translations and call SpeakTextAsync() for each.