Add input/output transcription to live audio example

Enable text transcription alongside audio streaming.
Output transcription prints inline, input transcription
in italic. Newlines separate input from output and after
sentence-ending punctuation.

Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>

Author: makelinux

command-line/node/main.mts

@@ -12,6 +12,8 @@ const model = 'gemini-3.1-flash-live-preview';
 const config = {
   responseModalities: [Modality.AUDIO],
   systemInstruction: "You are a helpful and friendly AI assistant.",
+  outputAudioTranscription: {},
+  inputAudioTranscription: {},
 };
 
 async function live() {
@@ -40,22 +42,36 @@ async function live() {
     process.stdin.pipe(speaker);
   }
 
+  let lastWasInput = false;
+
   async function messageLoop() {
     // Puts incoming messages in the audio queue.
     while (true) {
       const message = await waitMessage();
-      if (message.serverContent && message.serverContent.interrupted) {
+      const sc = message.serverContent;
+      if (!sc) continue;
+      if (sc.interrupted) {
         // Empty the queue on interruption to stop playback
         audioQueue.length = 0;
         continue;
       }
-      if (message.serverContent && message.serverContent.modelTurn && message.serverContent.modelTurn.parts) {
-        for (const part of message.serverContent.modelTurn.parts) {
-          if (part.inlineData && part.inlineData.data) {
+      if (sc.modelTurn?.parts) {
+        for (const part of sc.modelTurn.parts) {
+          if (part.inlineData?.data) {
             audioQueue.push(Buffer.from(part.inlineData.data, 'base64'));
           }
         }
       }
+      if (sc.outputTranscription?.text) {
+        if (lastWasInput) { process.stdout.write('\n'); lastWasInput = false; }
+        const t = sc.outputTranscription.text;
+        process.stdout.write(t);
+        if (/[.!?]\s*$/.test(t)) process.stdout.write('\n');
+      }
+      if (sc.inputTranscription?.text) {
+        if (!lastWasInput) { process.stdout.write('\n'); lastWasInput = true; }
+        process.stdout.write(`\x1b[3m${sc.inputTranscription.text}\x1b[0m`);
+      }
     }
   }
 

command-line/python/main.py

@@ -18,6 +18,8 @@
 CONFIG = {
     "response_modalities": ["AUDIO"],
     "system_instruction": "You are a helpful and friendly AI assistant.",
+    "output_audio_transcription": {},
+    "input_audio_transcription": {},
 }
 
 audio_queue_output = asyncio.Queue()
@@ -50,13 +52,33 @@ async def send_realtime(session):
 
 async def receive_audio(session):
     """Receives responses from GenAI and puts audio data into the speaker audio queue."""
+    last_was_input = False
     while True:
         turn = session.receive()
         async for response in turn:
-            if (response.server_content and response.server_content.model_turn):
-                for part in response.server_content.model_turn.parts:
+            sc = response.server_content
+            if not sc:
+                continue
+            if sc.model_turn:
+                for part in sc.model_turn.parts:
                     if part.inline_data and isinstance(part.inline_data.data, bytes):
                         audio_queue_output.put_nowait(part.inline_data.data)
+            if sc.output_transcription:
+                if last_was_input:
+                    print()
+                    last_was_input = False
+                t = sc.output_transcription.text
+                print(t, end="", flush=True)
+                if t.rstrip()[-1:] in '.!?':
+                    print()
+            if sc.input_transcription:
+                if not last_was_input:
+                    print()
+                    last_was_input = True
+                t = sc.input_transcription.text
+                print(f"\033[3m{t}\033[0m", end="", flush=True)
+                if t.rstrip()[-1:] in '.!?':
+                    print()
 
         # Empty the queue on interruption to stop playback
         while not audio_queue_output.empty():