Migrate to Gemini Live API

.env.example

@@ -1 +1 @@
-OPENAI_API_KEY=your_api_key
+GEMINI_API_KEY=your_api_key

Readme.md

@@ -1,27 +1,24 @@
-#  Speech Assistant with Twilio Voice and the OpenAI Realtime API (Python)
+#  Speech Assistant with Twilio Voice and the Gemini Live API (Python)
 
-This application demonstrates how to use Python, [Twilio Voice](https://www.twilio.com/docs/voice) and [Media Streams](https://www.twilio.com/docs/voice/media-streams), and [OpenAI's Realtime API](https://platform.openai.com/docs/) to make a phone call to speak with an AI Assistant. 
+This application demonstrates how to use Python, [Twilio Voice](https://www.twilio.com/docs/voice) and [Media Streams](https://www.twilio.com/docs/voice/media-streams), and [Google's Gemini Live API](https://ai.google.dev/gemini-api/docs/live) to make a phone call to speak with an AI Assistant.
 
-The application opens websockets with the OpenAI Realtime API and Twilio, and sends voice audio from one to the other to enable a two-way conversation.
+The application opens websockets with the Gemini Live API and Twilio, and sends voice audio from one to the other to enable a two-way conversation.
 
-See [here](https://www.twilio.com/en-us/blog/voice-ai-assistant-openai-realtime-api-python) for a tutorial overview of the code.
-
-This application uses the following Twilio products in conjunction with OpenAI's Realtime API:
+This application uses the following Twilio products in conjunction with the Gemini Live API:
 - Voice (and TwiML, Media Streams)
 - Phone Numbers
 
 > [!NOTE]
-> Outbound calling is beyond the scope of this app. However, we demoed [one way to do it here](https://www.twilio.com/en-us/blog/outbound-calls-python-openai-realtime-api-voice).
+> Outbound calling is beyond the scope of this app.
 
 ## Prerequisites
 
 To use the app, you will  need:
 
-- **Python 3.9+** We used \`3.9.13\` for development; download from [here](https://www.python.org/downloads/).
+- **Python 3.9+** (3.11 or earlier recommended; `audioop` is removed in 3.13+). Download from [here](https://www.python.org/downloads/).
 - **A Twilio account.** You can sign up for a free trial [here](https://www.twilio.com/try-twilio).
 - **A Twilio number with _Voice_ capabilities.** [Here are instructions](https://help.twilio.com/articles/223135247-How-to-Search-for-and-Buy-a-Twilio-Phone-Number-from-Console) to purchase a phone number.
-- **An OpenAI account and an OpenAI API Key.** You can sign up [here](https://platform.openai.com/).
-  - **OpenAI Realtime API access.**
+- **A Google AI Studio account and a Gemini API Key.** You can get one [here](https://aistudio.google.com/apikey).
 
 ## Local Setup
 
@@ -43,7 +40,7 @@ Once the tunnel has been opened, copy the `Forwarding` URL. It will look somethi
 need this when configuring your Twilio number setup.
 
 Note that the `ngrok` command above forwards to a development server running on port `5050`, which is the default port configured in this application. If
-you override the `PORT` defined in `index.js`, you will need to update the `ngrok` command accordingly.
+you override the `PORT` defined in `main.py`, you will need to update the `ngrok` command accordingly.
 
 Keep in mind that each time you run the `ngrok http` command, a new URL will be created, and you'll need to update it everywhere it is referenced below.
 
@@ -72,13 +69,13 @@ In your Phone Number configuration settings, update the first **A call comes in*
 
 ### Update the .env file
 
-Create a `/env` file, or copy the `.env.example` file to `.env`:
+Create a `.env` file, or copy the `.env.example` file to `.env`:
 
 ```
 cp .env.example .env
 ```
 
-In the .env file, update the `OPENAI_API_KEY` to your OpenAI API key from the **Prerequisites**.
+In the .env file, update the `GEMINI_API_KEY` to your Gemini API key from the **Prerequisites**.
 
 ## Run the app
 Once ngrok is running, dependencies are installed, Twilio is configured properly, and the `.env` is set up, run the dev server with the following command:
@@ -88,12 +85,13 @@ python main.py
 ## Test the app
 With the development server running, call the phone number you purchased in the **Prerequisites**. After the introduction, you should be able to talk to the AI Assistant. Have fun!
 
-## Special features
+## Audio Pipeline
+
+Twilio sends PCMU (G.711 μ-law, 8kHz) audio. Gemini Live expects raw PCM (16-bit, little-endian, 16kHz). The server handles transcoding:
 
-### Have the AI speak first
-To have the AI voice assistant talk before the user, uncomment the line `# await send_initial_conversation_item(openai_ws)`. The initial greeting is controlled in `async def send_initial_conversation_item(openai_ws)`.
+- **Twilio → Gemini**: PCMU 8kHz → `audioop.ulaw2lin` → PCM 8kHz → `audioop.ratecv` (upsample 8→16kHz) → Gemini
+- **Gemini → Twilio**: PCM 24kHz → struct averaging (downsample 24→8kHz) → `audioop.lin2ulaw` → PCMU → Twilio
 
-### Interrupt handling/AI preemption
-When the user speaks and OpenAI sends `input_audio_buffer.speech_started`, the code will clear the Twilio Media Streams buffer and send OpenAI `conversation.item.truncate`.
+## Interrupt handling / Barge-in
 
-Depending on your application's needs, you may want to use the [`input_audio_buffer.speech_stopped`](https://platform.openai.com/docs/api-reference/realtime-server-events/input-audio-buffer-speech-stopped) event, instead, or a combination of the two.
+Gemini Live handles barge-in natively with server-side VAD. When the user speaks during an AI response, Gemini sends `serverContent.interrupted: true`. The server then clears Twilio's audio buffer so the caller stops hearing the old response immediately. No manual truncation messages needed.