From d01efcb3084a0a3bd7302294d1072ab8ab247ecd Mon Sep 17 00:00:00 2001
From: AlexsanderHamir <alexsanderhamirgomesbaptista@gmail.com>
Date: Sat, 15 Nov 2025 15:58:41 -0800
Subject: [PATCH] speech set up

---
 no_cache_hits.py   |  48 +++++++++++++++++++++++++++++++++++++++++++++
 speech.mp3         | Bin 0 -> 104 bytes
 speech_config.yaml |   9 +++++++++
 3 files changed, 57 insertions(+)
 create mode 100644 no_cache_hits.py
 create mode 100644 speech.mp3
 create mode 100644 speech_config.yaml

diff --git a/no_cache_hits.py b/no_cache_hits.py
new file mode 100644
index 00000000000..1b3bf895f77
--- /dev/null
+++ b/no_cache_hits.py
@@ -0,0 +1,48 @@
+from locust import HttpUser, between, task
+
+
+class MyUser(HttpUser):
+    """
+    Minimal Locust user for repeatedly hitting `/v1/audio/speech`.
+    The goal is to measure server-side performance, so we avoid any extra work
+    (file writes, random generation, manual timing, custom event hooks, etc.)
+    that could inflate client-side latency.
+    """
+
+    wait_time = between(0.5, 1)
+    host = "http://0.0.0.0:8090"
+
+    def on_start(self):
+        self.api_key = "sk-1234"
+        self.model_name = "fake-openai-speech"
+        self.headers = {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json",
+        }
+        self.prompt_counter = 0
+
+    @task
+    def audio_speech_request(self):
+        self.prompt_counter += 1
+        # Ensure prompts differ slightly so the backend can't reuse cached audio.
+        prompt = (
+            "Generate a short spoken status update mentioning counter "
+            f"{self.prompt_counter}."
+        )
+
+        response = self.client.post(
+            "v1/audio/speech",
+            json={
+                "model": self.model_name,
+                "input": prompt,
+                "voice": "alloy",
+                "format": "mp3",
+            },
+            headers=self.headers,
+            name="audio_speech",
+        )
+
+        if response.status_code != 200:
+            # log the errors in error.txt
+            with open("error.txt", "a") as error_log:
+                error_log.write(response.text + "\n")
\ No newline at end of file
diff --git a/speech.mp3 b/speech.mp3
new file mode 100644
index 0000000000000000000000000000000000000000..f4f854d9bd215e2493d48b4bc4d39804bd79c038
GIT binary patch
literal 104
NcmezWdjbPJ000JT0*e3u

literal 0
HcmV?d00001

diff --git a/speech_config.yaml b/speech_config.yaml
new file mode 100644
index 00000000000..ad9920a2793
--- /dev/null
+++ b/speech_config.yaml
@@ -0,0 +1,9 @@
+model_list:
+  - model_name: fake-openai-speech
+    litellm_params:
+      model: openai/gpt-4o-mini-tts
+      api_base: http://0.0.0.0:8090/
+      api_key: sk-1234
+    model_info:
+      mode: audio_speech
+  
\ No newline at end of file