Generative AI Speech

Note

“Generative AI Speech” is jointly developed by RTKSG SD3 and ChungYi Fu (Kaohsiung, Taiwan)

Special thanks and credits to the efforts and contributions for all developers.

Materials

• AMB82-mini x 1

• MicroSD card

• PushButton x1

• 220 ohm resistor x1

Example

GenAISpeech_Whisper

In this example, we will be using Ameba Pro2 development board to record and send audio recording to LLM server for voice transcription or translation.

Open Generative AI Speech Whisper example in “File” -> “Examples” -> “AmebaNN” -> “MultimediaAI” -> “GenAISpeech_Whisper”.

In the highlighted code snippet, fill in the “ssid” with your WiFi network SSID and “pass” with the network password.

Choose your API server and and paste your API key accordingly.

Modify the api_path according to your server and tasks (transcription or translation).

Modify the audio_model to select the LLM model which suits your application needs.

You may modify the filename and recording duration here.

Connect the pushbutton and resistor to AMB82 Mini as shown below.

Compile and run the example.

Open the serial monitor to view the logs.

Press button for 2s and wait for the green LED to light up, then speak into the microphone within the pre-defined recording duration.

Transcription or translation of the audio file will be printed out on serial monitor.

GenAISpeech_Gemini

Open Generative AI Speech Gemini example in “File” -> “Examples” -> “AmebaNN” -> “MultimediaAI” -> “GenAISpeech_Gemini”.

Fill in the “ssid” with your WiFi network SSID and “pass” with the network password.

Then, fill in your Gemini api key.

You may also modify the recording duration and filename.

Connect the pushbutton and resistor to AMB82 Mini as shown below.

Compile and run the example.

Open the serial monitor to view the logs.

Press button once, recording will start after 3 seconds of blue LED blinking, then speak into the microphone within the pre-defined recording duration.

Response from Gemini will be printed out on serial monitor.

GenAISpeech_Gemini_LEDControl

Open Generative AI Speech Gemini example in “File” -> “Examples” -> “AmebaNN” -> “MultimediaAI” -> “GenAISpeech_Gemini_LEDControl”.

Fill in the “ssid” with your WiFi network SSID and “pass” with the network password.

Then, fill in your Gemini api key.

Connect the pushbutton and resistor to AMB82 Mini as shown below.

Compile and run the example.

Open the serial monitor to view the logs.

Press button once, recording will start after 3 seconds of blue LED blinking, then speak “LED ON” into the microphone within the pre-defined recording duration.

The green LED will be on subsequently.

You may try the vice versa as well.

Online LLM Models

Various online servers and LLM models featured in the SDK:

Host	Transcription Endpoint	Translation Endpoint	Model	Rate Limit	Pricing
api.openai.com	/v1/audio/transcriptions	/v1/audio/translations	whisper-1	500 RPM	Chargeable (Tier 1)
api.groq.com	/openai/v1/audio/transcriptions	/openai/v1/audio/translations	whisper-large-v3-turbo or whisper-large-v3	20 RPM	Free of charge
generativelanguage.googleapis.com	/v1beta/models/<model>		gemini-2.0-flash	15 RPM	Free of charge

Rate Limit References

openAI: https://platform.openai.com/docs/guides/rate-limits?context=tier-one#usage-tiers

GroqCloud: https://console.groq.com/settings/limits

Google AI Studio: https://ai.google.dev/gemini-api/docs/rate-limits

Resources

openAI platform

https://platform.openai.com/docs/guides/speech-to-text

GroqCloud

https://console.groq.com/docs/speech-text

Google AI Studio

https://ai.google.dev/gemini-api/docs/audio?lang=rest