Audio init

app.py

@@ -1,6 +1,10 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 from transformers import pipeline
+from scipy.io.wavfile import write as write_wav
+
+AUDIO_FILE_PATH = "bark_generation.wav"
+synthesizer = pipeline("text-to-speech", "suno/bark-small")
 
 """
 For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
@@ -8,65 +12,36 @@ For more information on `huggingface_hub` Inference API support, please check th
 
 client = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")
 
-tts = pipeline("text-to-speech", model="facebook/tts_transformer-EN-GLACIAL")
-
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    # TODO: Doesn't stream ATM
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-        response += token
-        yield response
-
-    audio = tts(response)
-    return response, audio
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-    outputs=[gr.Textbox(), gr.Audio()],
-)
-
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot(type="messages")
+    audio_box = gr.Audio(autoplay=True)
+    msg = gr.Textbox(submit_btn=True)
+    clear = gr.Button("Clear")
+
+    def synthesize_audio(text):
+        speech = synthesizer(text, forward_params={"do_sample": True})
+        write_wav(AUDIO_FILE_PATH, rate=speech["sampling_rate"], data=speech["audio"])
+
+    def user(user_message, history: list):
+        return "", history + [{"role": "user", "content": user_message}]
+
+    def bot(history: list):
+        history.append({"role": "assistant", "content": ""})
+        for message in client.chat_completion(
+            history,
+            stream=True,
+        ):
+            token = message.choices[0].delta.content
+            history[-1]["content"] += token
+            yield history, None
+
+        synthesize_audio(history[-1]["content"])
+        return history, AUDIO_FILE_PATH
+
+    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
+        bot, chatbot, [chatbot, audio_box]
+    )
+    clear.click(lambda: None, None, chatbot, queue=False)
 
 if __name__ == "__main__":
     demo.launch()

requirements.txt

@@ -1,4 +1,5 @@
 huggingface_hub
 unsloth
 gradio
+scipy
 transformers