diff --git a/src/zai/api_resource/audio/audio.py b/src/zai/api_resource/audio/audio.py
index 2de7b47..a3cd152 100644
--- a/src/zai/api_resource/audio/audio.py
+++ b/src/zai/api_resource/audio/audio.py
@@ -3,7 +3,6 @@
 from typing import TYPE_CHECKING, Mapping, Optional, cast
 
 import httpx
-from httpx import stream
 
 from zai.core import (
 	NOT_GIVEN,
@@ -23,6 +22,8 @@
 from zai.types.sensitive_word_check import SensitiveWordCheckRequest
 
 from .transcriptions import Transcriptions
+from zai.core._streaming import StreamResponse
+from zai.types.audio import AudioSpeechChunk
 
 if TYPE_CHECKING:
 	from zai._client import ZaiClient
@@ -60,7 +61,7 @@ def speech(
 		speed: float | None = 1.0,
 		volume: float | None = 1.0,
 		stream: bool | None = False
-	) -> HttpxBinaryResponseContent:
+	) -> HttpxBinaryResponseContent | StreamResponse[AudioSpeechChunk]:
 		"""
 		Generate speech audio from text input
 
@@ -83,7 +84,6 @@ def speech(
 				'voice': voice,
 				'response_format': response_format,
 				'encode_format': encode_format,
-				'sensitive_word_check': sensitive_word_check,
 				'request_id': request_id,
 				'user_id': user_id,
 				'speed': speed,
@@ -96,6 +96,8 @@ def speech(
 			body=maybe_transform(body, AudioSpeechParams),
 			options=make_request_options(extra_headers=extra_headers, extra_body=extra_body, timeout=timeout),
 			cast_type=HttpxBinaryResponseContent,
+			stream=stream or False,
+			stream_cls=StreamResponse[AudioSpeechChunk]
 		)
 
 	def customization(
diff --git a/src/zai/types/audio/__init__.py b/src/zai/types/audio/__init__.py
index ad73c7b..f53dd5e 100644
--- a/src/zai/types/audio/__init__.py
+++ b/src/zai/types/audio/__init__.py
@@ -1,5 +1,6 @@
 from .audio_customization_param import AudioCustomizationParam
+from .audio_speech_chunk import AudioSpeechChunk
 from .audio_speech_params import AudioSpeechParams
 from .transcriptions_create_param import TranscriptionsParam
 
-__all__ = ['AudioSpeechParams', 'AudioCustomizationParam', 'TranscriptionsParam']
+__all__ = ['AudioSpeechParams', 'AudioCustomizationParam', 'TranscriptionsParam', 'AudioSpeechChunk']
diff --git a/src/zai/types/audio/audio_speech_chunk.py b/src/zai/types/audio/audio_speech_chunk.py
new file mode 100644
index 0000000..3d18c3b
--- /dev/null
+++ b/src/zai/types/audio/audio_speech_chunk.py
@@ -0,0 +1,32 @@
+from typing import List, Optional, Dict, Any
+
+from ...core import BaseModel
+
+__all__ = [
+    "AudioSpeechChunk",
+    "AudioError",
+    "AudioSpeechChoice",
+    "AudioSpeechDelta"
+]
+
+
+class AudioSpeechDelta(BaseModel):
+    content: Optional[str] = None
+    role: Optional[str] = None
+
+
+class AudioSpeechChoice(BaseModel):
+    delta: AudioSpeechDelta
+    finish_reason: Optional[str] = None
+    index: int
+
+class AudioError(BaseModel):
+    code: Optional[str] = None
+    message: Optional[str] = None
+
+
+class AudioSpeechChunk(BaseModel):
+    choices: List[AudioSpeechChoice]
+    request_id: Optional[str] = None
+    created: Optional[int] = None
+    error: Optional[AudioError] = None
\ No newline at end of file
diff --git a/src/zai/types/audio/audio_speech_params.py b/src/zai/types/audio/audio_speech_params.py
index c8eb538..8843ab3 100644
--- a/src/zai/types/audio/audio_speech_params.py
+++ b/src/zai/types/audio/audio_speech_params.py
@@ -29,3 +29,7 @@ class AudioSpeechParams(TypedDict, total=False):
 	sensitive_word_check: Optional[SensitiveWordCheckRequest]
 	request_id: str
 	user_id: str
+	encode_format: str
+	speed: float
+	volume: float
+	stream: bool
diff --git a/tests/integration_tests/test_audio.py b/tests/integration_tests/test_audio.py
index e9ef33c..3dbcf66 100644
--- a/tests/integration_tests/test_audio.py
+++ b/tests/integration_tests/test_audio.py
@@ -1,3 +1,4 @@
+import base64
 import logging
 import logging.config
 from pathlib import Path
@@ -5,7 +6,6 @@
 import zai
 from zai import ZaiClient
 
-
 def test_audio_speech(logging_conf):
 	logging.config.dictConfig(logging_conf)  # type: ignore
 	client = ZaiClient()  # Fill in your own API Key
@@ -17,11 +17,19 @@ def test_audio_speech(logging_conf):
 			voice='female',
 			response_format='pcm',
 			encode_format='base64',
-			stream=False,
+			stream=True,
 			speed=1.0,
 			volume=1.0,
 		)
-		response.stream_to_file(speech_file_path)
+		with open("output.pcm", "wb") as f:
+			for item in response:
+				choice = item.choices[0]
+				index = choice.index
+				finish_reason = choice.finish_reason
+				if choice.delta is None:
+					break
+				audio_delta = choice.delta.content
+				f.write(base64.b64decode(audio_delta))
 
 	except zai.core._errors.APIRequestFailedError as err:
 		print(err)