diff --git a/src/rook/tts.py b/src/rook/tts.py
new file mode 100644
index 0000000..7e248e6
--- /dev/null
+++ b/src/rook/tts.py
@@ -0,0 +1,50 @@
+import os
+from typing import Optional
+
+
+class TTSError(Exception):
+ pass
+
+
+async def speak(text: str) -> None:
+ el_key = os.environ.get('ELEVENLABS_API_KEY')
+ g_key = os.environ.get('GOOGLE_TTS_KEY')
+ audio: Optional[bytes] = None
+ if el_key:
+  try:
+   audio = _speak_elevenlabs(text)
+  except Exception:
+   if g_key:
+    audio = _speak_google(text)
+ elif g_key:
+  audio = _speak_google(text)
+ if audio is None:
+  raise TTSError('No TTS keys configured')
+ _play_audio(audio)
+
+
+def _speak_elevenlabs(text: str) -> bytes:
+ from elevenlabs import ElevenLabs
+ client = ElevenLabs(api_key=os.environ['ELEVENLABS_API_KEY'])
+ audio_gen = client.generate(text=text, voice=os.environ.get('ELEVENLABS_VOICE_ID', ''))
+ return b''.join(audio_gen)
+
+
+def _speak_google(text: str) -> bytes:
+ from google.cloud import texttospeech
+ client = texttospeech.TextToSpeechClient()
+ response = client.synthesize_speech(
+  input=texttospeech.SynthesisInput(text=text),
+  voice=texttospeech.VoiceSelectionParams(
+   language_code='en-US',
+   ssml_gender=texttospeech.SsmlVoiceGender.NEUTRAL,
+  ),
+  audio_config=texttospeech.AudioConfig(
+   audio_encoding=texttospeech.AudioEncoding.MP3,
+  ),
+ )
+ return response.audio_content
+
+
+def _play_audio(audio_bytes: bytes) -> None:
+ pass
diff --git a/tests/test_tts.py b/tests/test_tts.py
new file mode 100644
index 0000000..d47f4c3
--- /dev/null
+++ b/tests/test_tts.py
@@ -0,0 +1,90 @@
+import asyncio
+import inspect
+import pytest
+from unittest.mock import patch, MagicMock
+
+
+def test_speak_elevenlabs_called_when_key_set():
+ from rook.tts import speak
+ env = {'ELEVENLABS_API_KEY': 'key', 'ELEVENLABS_VOICE_ID': 'vid'}
+ with patch.dict('os.environ', env, clear=True):
+  with patch('rook.tts._speak_elevenlabs', return_value=b'audio') as mock_el:
+   with patch('rook.tts._play_audio'):
+    asyncio.run(speak('hello'))
+    mock_el.assert_called_with('hello')
+
+
+def test_speak_falls_back_to_google_on_elevenlabs_error():
+ from rook.tts import speak
+ env = {
+  'ELEVENLABS_API_KEY': 'key',
+  'ELEVENLABS_VOICE_ID': 'vid',
+  'GOOGLE_TTS_KEY': 'gkey',
+ }
+ with patch.dict('os.environ', env, clear=True):
+  with patch('rook.tts._speak_elevenlabs', side_effect=Exception('API error')):
+   with patch('rook.tts._speak_google', return_value=b'audio') as mock_google:
+    with patch('rook.tts._play_audio'):
+     asyncio.run(speak('hello'))
+     mock_google.assert_called()
+
+
+def test_speak_google_only_when_no_elevenlabs():
+ from rook.tts import speak
+ env = {'GOOGLE_TTS_KEY': 'gkey'}
+ with patch.dict('os.environ', env, clear=True):
+  with patch('rook.tts._speak_google', return_value=b'audio') as mock_google:
+   with patch('rook.tts._play_audio'):
+    asyncio.run(speak('hi'))
+    mock_google.assert_called_with('hi')
+
+
+def test_speak_raises_tts_error_when_no_keys():
+ from rook.tts import speak, TTSError
+ with patch.dict('os.environ', {}, clear=True):
+  with pytest.raises(TTSError):
+   asyncio.run(speak('hi'))
+
+
+def test_speak_calls_play_audio():
+ from rook.tts import speak
+ env = {'ELEVENLABS_API_KEY': 'key', 'ELEVENLABS_VOICE_ID': 'vid'}
+ with patch.dict('os.environ', env, clear=True):
+  with patch('rook.tts._speak_elevenlabs', return_value=b'bytes'):
+   with patch('rook.tts._play_audio') as mock_play:
+    asyncio.run(speak('test'))
+    mock_play.assert_called_with(b'bytes')
+
+
+def test_speak_is_coroutine():
+ from rook.tts import speak
+ assert inspect.iscoroutinefunction(speak)
+
+
+def test_speak_elevenlabs_internal():
+ mock_client = MagicMock()
+ mock_client.generate.return_value = iter([b'chunk1', b'chunk2'])
+ mock_elevenlabs_cls = MagicMock(return_value=mock_client)
+ with patch('elevenlabs.ElevenLabs', mock_elevenlabs_cls):
+  with patch.dict('os.environ', {'ELEVENLABS_API_KEY': 'k', 'ELEVENLABS_VOICE_ID': 'v'}):
+   from rook.tts import _speak_elevenlabs
+   result = _speak_elevenlabs('hi')
+   assert result == b'chunk1chunk2'
+
+
+def test_speak_google_internal():
+ mock_response = MagicMock()
+ mock_response.audio_content = b'speech'
+ mock_client = MagicMock()
+ mock_client.synthesize_speech.return_value = mock_response
+ mock_tts_cls = MagicMock(return_value=mock_client)
+ with patch('google.cloud.texttospeech.TextToSpeechClient', mock_tts_cls):
+  from rook.tts import _speak_google
+  result = _speak_google('hi')
+  assert result == b'speech'
+
+
+def test_play_audio_noop():
+ from rook.tts import _play_audio
+ result = _play_audio(b'data')
+ assert result is None