Validate the voice translation pipeline

Test the complete workflow

You’ll validate the application that you built by checking its English transcript, Spanish translation, and spoken output. You’ll then test a new recording with the Mac disconnected from the network to confirm offline operation.

You’ll check the transcript first because a recognition error changes the text Gemma receives and can affect the translation and speech.

Keep app.py running and Translate to set to Spanish. Use the browser to check each result in sequence.

Record the following English sentence in a quiet environment:

    

        
        
Could you show me the way to the city center?

    

Select Translate and verify the results in this order:

1. Transcript contains the sentence you spoke.
2. The translation appears progressively in Spanish, preserves the meaning of the sentence, and contains no explanation or formatting.
3. The status changes to Synthesizing… after translation finishes.
4. Translated speech appears and plays the complete translated sentence with Spanish pronunciation.
5. The status returns to Ready.

A suitable Spanish translation is “¿Podría indicarme el camino al centro de la ciudad?” Minor differences in punctuation or wording are expected.

Verify on-device operation

Keep the browser tab and app.py running, then temporarily disconnect the Mac from Wi-Fi or Ethernet. Record another English sentence and translate it into Spanish.

The transcript, translation, and speech should still complete because all model files and inference runtimes are local. Reconnect the Mac after the test.

Stop the application

Return to the terminal running the application and press Ctrl+C. The finally block closes the active conversation, and the engine context releases the Gemma resources.

What you’ve accomplished

You’ve connected and validated English speech recognition, English-to-Spanish translation with Gemma 4 E2B, and Spanish speech synthesis with Supertonic 3 in a local Python application.

To try another voice or adjust speech generation, see the Sherpa-ONNX Supertonic documentation . Test your changes with tts.py before running the full application with app.py.