diff --git a/.gitignore b/.gitignore
index 64067ad..ee4effd 100644
--- a/.gitignore
+++ b/.gitignore
@@ -127,6 +127,7 @@ pypy/
 pypy*
 cpython*/
 venv
+.python-version
 
 # Installer logs
 pip-log.txt
diff --git a/src/icespeak/settings.py b/src/icespeak/settings.py
index 9bef80b..67d67e8 100644
--- a/src/icespeak/settings.py
+++ b/src/icespeak/settings.py
@@ -182,6 +182,19 @@ class Keys(BaseModel):
     # TODO: Re-implement TTS with Tiro
     tiro: Literal[None] = Field(default=None)
 
+    def __hash__(self):
+        return hash((self.azure, self.aws, self.google, self.tiro))
+
+    def __eq__(self, other):
+        if isinstance(other, Keys):
+            return (self.azure, self.aws, self.google, self.tiro) == (
+                other.azure,
+                other.aws,
+                other.google,
+                other.tiro,
+            )
+        return False
+
 
 API_KEYS = Keys()
 
diff --git a/src/icespeak/tts.py b/src/icespeak/tts.py
index d17cd0b..ef95f1e 100644
--- a/src/icespeak/tts.py
+++ b/src/icespeak/tts.py
@@ -34,7 +34,7 @@
 
 from cachetools import LFUCache, cached
 
-from .settings import SETTINGS, TRACE
+from .settings import SETTINGS, TRACE, Keys
 from .transcribe import TranscriptionOptions
 
 # TODO: Re implement Tiro
@@ -64,7 +64,7 @@ def _setup_voices() -> tuple[VoicesT, ServicesT]:
     for service in services:
         _LOG.debug("Loading voices from service: %s", service)
         if not service.available:
-            _LOG.info("Voices from service %s not availble.", service)
+            _LOG.info("Voices from service %s not available.", service)
             continue
         for voice, info in service.voices.items():
             # Info about each voice
@@ -153,6 +153,7 @@ def tts_to_file(
     transcription_options: TranscriptionOptions | None = None,
     *,
     transcribe: bool = True,
+    keys_override: Keys | None = None,
 ) -> TTSOutput:
     """
     # Text-to-speech
@@ -195,7 +196,7 @@ def tts_to_file(
         text = service.Transcriber.token_transcribe(text, options=transcription_options)
 
     output = TTSOutput(
-        file=service.text_to_speech(text, tts_options),
+        file=service.text_to_speech(text, tts_options, keys_override),
         text=text,
     )
     _LOG.debug("tts_to_file, out: %s", output)
diff --git a/src/icespeak/voices/__init__.py b/src/icespeak/voices/__init__.py
index 2565fec..bd5c73d 100644
--- a/src/icespeak/voices/__init__.py
+++ b/src/icespeak/voices/__init__.py
@@ -28,7 +28,7 @@
 
 from pydantic import BaseModel, Field
 
-from icespeak.settings import MAX_SPEED, MIN_SPEED, SETTINGS, TextFormats
+from icespeak.settings import MAX_SPEED, MIN_SPEED, SETTINGS, Keys, TextFormats
 from icespeak.transcribe import DefaultTranscriber
 
 if TYPE_CHECKING:
@@ -125,5 +125,7 @@ def load_api_keys(self) -> None:
         raise NotImplementedError
 
     @abstractmethod
-    def text_to_speech(self, text: str, options: TTSOptions) -> Path:
+    def text_to_speech(
+        self, text: str, options: TTSOptions, keys_override: Keys | None = None
+    ) -> Path:
         raise NotImplementedError
diff --git a/src/icespeak/voices/aws_polly.py b/src/icespeak/voices/aws_polly.py
index 4de112c..f113929 100644
--- a/src/icespeak/voices/aws_polly.py
+++ b/src/icespeak/voices/aws_polly.py
@@ -30,7 +30,7 @@
 
 import boto3
 
-from icespeak.settings import API_KEYS, SETTINGS
+from icespeak.settings import API_KEYS, SETTINGS, AWSPollyKey, Keys
 
 from . import BaseVoice, ModuleAudioFormatsT, ModuleVoicesT, TTSOptions
 
@@ -47,6 +47,14 @@ class AWSPollyVoice(BaseVoice):
 
     _lock = Lock()
 
+    def _create_client(self, aws_key: AWSPollyKey) -> boto3.client:
+        return boto3.client(
+            "polly",
+            region_name=aws_key.region_name.get_secret_value(),
+            aws_access_key_id=aws_key.aws_access_key_id.get_secret_value(),
+            aws_secret_access_key=aws_key.aws_secret_access_key.get_secret_value(),
+        )
+
     @property
     @override
     def name(self):
@@ -69,16 +77,18 @@ def load_api_keys(self):
         self._aws_client: Any = None
         with AWSPollyVoice._lock:
             if self._aws_client is None:
-                # See boto3.Session.client for arguments
-                self._aws_client = boto3.client(
-                    "polly",
-                    region_name=API_KEYS.aws.region_name.get_secret_value(),
-                    aws_access_key_id=API_KEYS.aws.aws_access_key_id.get_secret_value(),
-                    aws_secret_access_key=API_KEYS.aws.aws_secret_access_key.get_secret_value(),
-                )
+                self._aws_client = self._create_client(API_KEYS.aws)
 
     @override
-    def text_to_speech(self, text: str, options: TTSOptions):
+    def text_to_speech(
+        self, text: str, options: TTSOptions, keys_override: Keys | None = None
+    ):
+        if keys_override and keys_override.aws:
+            _LOG.debug("Using overridden AWS keys")
+            client = self._create_client(keys_override.aws)
+        else:
+            _LOG.debug("Using default AWS keys")
+            client = self._aws_client
         # Special preprocessing for SSML markup
         if options.text_format == "ssml":
             # Adjust voice speed as appropriate
@@ -99,7 +109,7 @@ def text_to_speech(self, text: str, options: TTSOptions):
                 "OutputFormat": options.audio_format,
             }
             _LOG.debug("Synthesizing with AWS Polly: %s", aws_args)
-            response: dict[str, Any] = self._aws_client.synthesize_speech(**aws_args)
+            response: dict[str, Any] = client.synthesize_speech(**aws_args)
         except Exception:
             _LOG.exception("Error synthesizing speech.")
             raise
diff --git a/src/icespeak/voices/azure.py b/src/icespeak/voices/azure.py
index 63b173b..4e1c882 100644
--- a/src/icespeak/voices/azure.py
+++ b/src/icespeak/voices/azure.py
@@ -29,7 +29,7 @@
 
 import azure.cognitiveservices.speech as speechsdk
 
-from icespeak.settings import API_KEYS, SETTINGS
+from icespeak.settings import API_KEYS, SETTINGS, Keys
 from icespeak.transcribe import DefaultTranscriber, strip_markup
 
 from . import BaseVoice, ModuleAudioFormatsT, ModuleVoicesT, TTSOptions
@@ -179,10 +179,18 @@ def load_api_keys(self):
         AzureVoice.AZURE_REGION = API_KEYS.azure.region.get_secret_value()
 
     @override
-    def text_to_speech(self, text: str, options: TTSOptions):
-        speech_conf = speechsdk.SpeechConfig(
-            subscription=AzureVoice.AZURE_KEY, region=AzureVoice.AZURE_REGION
-        )
+    def text_to_speech(
+        self, text: str, options: TTSOptions, keys_override: Keys | None = None
+    ):
+        if keys_override and keys_override.azure:
+            _LOG.debug("Using overridden Azure keys")
+            subscription = keys_override.azure.key.get_secret_value()
+            region = keys_override.azure.region.get_secret_value()
+        else:
+            _LOG.debug("Using default Azure keys")
+            subscription = AzureVoice.AZURE_KEY
+            region = AzureVoice.AZURE_REGION
+        speech_conf = speechsdk.SpeechConfig(subscription=subscription, region=region)
 
         azure_voice_id = AzureVoice._VOICES[options.voice]["id"]
         speech_conf.speech_synthesis_voice_name = azure_voice_id
diff --git a/src/icespeak/voices/tiro.py b/src/icespeak/voices/tiro.py
index 5bc2366..7811c13 100644
--- a/src/icespeak/voices/tiro.py
+++ b/src/icespeak/voices/tiro.py
@@ -29,7 +29,7 @@
 
 import requests
 
-from icespeak.settings import SETTINGS
+from icespeak.settings import SETTINGS, Keys
 from icespeak.transcribe import strip_markup
 
 from . import BaseVoice, ModuleAudioFormatsT, ModuleVoicesT, TTSOptions
@@ -70,7 +70,9 @@ def load_api_keys(self):
         pass
 
     @override
-    def text_to_speech(self, text: str, options: TTSOptions):
+    def text_to_speech(
+        self, text: str, options: TTSOptions, keys_override: Keys | None = None
+    ):
         # TODO: Tiro's API supports a subset of SSML tags
         # See https://tts.tiro.is/#tag/speech/paths/~1v0~1speech/post
 
diff --git a/tests/test_tts.py b/tests/test_tts.py
index 0e56f8c..83286bd 100644
--- a/tests/test_tts.py
+++ b/tests/test_tts.py
@@ -18,13 +18,23 @@
 
 
 """
+# ruff: noqa: S106
 from __future__ import annotations
 
+from unittest.mock import MagicMock, patch
+
 import pytest
 
 from icespeak import TTSOptions, tts_to_file
-from icespeak.settings import API_KEYS, TextFormats, suffix_for_audiofmt
+from icespeak.settings import (
+    API_KEYS,
+    AWSPollyKey,
+    Keys,
+    TextFormats,
+    suffix_for_audiofmt,
+)
 from icespeak.transcribe import strip_markup
+from icespeak.tts import SERVICES, VOICES
 
 
 def test_voices_utils():
@@ -57,6 +67,20 @@ def test_AWSPolly_speech_synthesis():
     path.unlink()
 
 
+@pytest.mark.skipif(API_KEYS.aws is None, reason="Missing AWS Polly API Key.")
+@pytest.mark.network()
+def test_AWSPolly_speech_synthesis_with_keys_override():
+    tts_out = tts_to_file(
+        _TEXT,
+        TTSOptions(text_format=TextFormats.TEXT, audio_format="mp3", voice="Dora"),
+        keys_override=API_KEYS,
+    )
+    path = tts_out.file
+    assert path.is_file(), "Expected audio file to exist"
+    assert path.stat().st_size > _MIN_AUDIO_SIZE, "Expected longer audio data"
+    path.unlink()
+
+
 @pytest.mark.skipif(API_KEYS.azure is None, reason="Missing Azure API Key.")
 @pytest.mark.network()
 def test_Azure_speech_synthesis():
@@ -71,6 +95,21 @@ def test_Azure_speech_synthesis():
     path.unlink()
 
 
+@pytest.mark.skipif(API_KEYS.azure is None, reason="Missing Azure API Key.")
+@pytest.mark.network()
+def test_Azure_speech_synthesis_with_keys_override():
+    # Test Azure Cognitive Services
+    tts_out = tts_to_file(
+        _TEXT,
+        TTSOptions(text_format=TextFormats.TEXT, audio_format="mp3", voice="Gudrun"),
+        keys_override=API_KEYS,
+    )
+    path = tts_out.file
+    assert path.is_file(), "Expected audio file to exist"
+    assert path.stat().st_size > _MIN_AUDIO_SIZE, "Expected longer audio data"
+    path.unlink()
+
+
 @pytest.mark.skipif(API_KEYS.google is None, reason="Missing Google API Key.")
 @pytest.mark.network()
 def test_Google_speech_synthesis():
@@ -97,3 +136,30 @@ def test_Tiro_speech_synthesis():
     assert path.is_file(), "Expected audio file to exist"
     assert path.stat().st_size > _MIN_AUDIO_SIZE, "Expected longer audio data"
     path.unlink()
+
+
+@patch.dict(SERVICES, {"mock_service": MagicMock()})
+@patch.dict(VOICES, {"Dora": {"service": "mock_service"}})
+def test_keys_override_in_tts_to_file():
+    """Test if keys_override is correctly passed into service.text_to_speech."""
+    _TEXT = "Test"
+    SERVICES["mock_service"].audio_formats = ["mp3"]
+    keys_override = Keys(
+        aws=AWSPollyKey(
+            aws_access_key_id="test",
+            aws_secret_access_key="test",
+            region_name="test",
+        )
+    )
+    opts = TTSOptions(text_format=TextFormats.TEXT, audio_format="mp3", voice="Dora")
+    tts_to_file(
+        _TEXT,
+        opts,
+        transcribe=False,
+        keys_override=keys_override,
+    )
+    SERVICES["mock_service"].text_to_speech.assert_called_once_with(
+        _TEXT,
+        opts,
+        keys_override,
+    )