This {component_name.lower()} is based on
MERaLiON-AudioLLM,
developed by I2R, A*STAR, in collaboration with AISG, Singapore.
{description}
""",
unsafe_allow_html=True
)
st.markdown(
f"""
This {component_name.lower()} is based on
MERaLiON-AudioLLM.{concise_description}
""",
unsafe_allow_html=True
)
@st.fragment
def sidebar_fragment():
with st.container(height=256, border=False):
st.page_link("pages/playground.py", disabled=st.session_state.disprompt, label="🚀 Playground")
st.page_link("pages/agent.py", disabled=st.session_state.disprompt, label="👥 Cascade System")
st.page_link("pages/voice_chat.py", disabled=st.session_state.disprompt, label="🗣️ End-to-End Voice Chat")
st.divider()
st.slider(label='Temperature', min_value=0.0, max_value=2.0, value=0.1, key='temperature')
st.slider(label='Top P', min_value=0.0, max_value=1.0, value=0.9, key='top_p')
st.slider(label="Repetition Penalty", min_value=1.0, max_value=1.2, value=1.1, key="repetition_penalty")
@st.fragment
def successful_example_section(audio_sample_names, audio_array_state, audio_base64_state, restore_state={}):
st.markdown(":fire: **Successful Tasks and Examples**")
sample_name = st.selectbox(
label="**Select Audio:**",
label_visibility="collapsed",
options=audio_sample_names,
format_func=lambda o: AUDIO_SAMPLES_W_INSTRUCT[o]["apperance"],
index=None,
placeholder="Select an audio sample:",
on_change=lambda: st.session_state.update(
on_select=True,
disprompt=True,
**copy.deepcopy(restore_state)
),
key='select')
if sample_name and st.session_state.on_select:
file_name = sample_name.split("#")[0]
audio_bytes = open(f"audio_samples/{file_name}.wav", "rb").read()
st.session_state.update(
on_select=False,
new_prompt=AUDIO_SAMPLES_W_INSTRUCT[sample_name]["instructions"][0]
)
st.session_state[audio_array_state], st.session_state[audio_base64_state] = \
process_audio_bytes(audio_bytes)
st.rerun(scope="app")
@st.dialog("Specify audio context for analysis")
def audio_attach_dialogue(audio_array_state, audio_base64_state, restore_state={}):
st.markdown("**Upload**")
uploaded_file = st.file_uploader(
label="**Upload Audio:**",
label_visibility="collapsed",
type=['wav', 'mp3'],
on_change=lambda: st.session_state.update(
on_upload=True,
**copy.deepcopy(restore_state)
),
key='upload'
)
if uploaded_file and st.session_state.on_upload:
audio_bytes = uploaded_file.read()
st.session_state[audio_array_state], st.session_state[audio_base64_state] = \
process_audio_bytes(audio_bytes)
st.session_state.on_upload = False
st.rerun()
st.markdown("**Record**")
uploaded_file = st.audio_input(
label="**Record Audio:**",
label_visibility="collapsed",
on_change=lambda: st.session_state.update(
on_record=True,
**copy.deepcopy(restore_state)
),
key='record'
)
if uploaded_file and st.session_state.on_record:
audio_bytes = uploaded_file.read()
st.session_state[audio_array_state], st.session_state[audio_base64_state] = \
process_audio_bytes(audio_bytes)
st.session_state.on_record = False
st.rerun()
def retrive_response_with_ui(
model_name: str,
text_input: str,
array_audio_input: np.ndarray,
base64_audio_input: str,
prefix: str = "",
stream: bool = True,
normalise_response: bool = False,
history: Optional[List] = None,
show_warning: bool = True,
**kwargs
):
if history is None:
history = []
# Prepare request data
request_data = {
"text_input": str(text_input),
"model_name": str(model_name),
"array_audio_input": array_audio_input.tolist(), # Convert numpy array to list
"base64_audio_input": str(base64_audio_input) if base64_audio_input else None,
"history": list(history) if history else None,
"stream": bool(stream),
"max_completion_tokens": int(st.session_state.max_completion_tokens),
"temperature": float(st.session_state.temperature),
"top_p": float(st.session_state.top_p),
"repetition_penalty": float(st.session_state.repetition_penalty),
"top_k": int(st.session_state.top_k),
"length_penalty": float(st.session_state.length_penalty),
"seed": int(st.session_state.seed),
"extra_params": {}
}
# print(request_data)
# print(model_name)
error_msg = ""
warnings = []
response = ""
try:
if stream:
# Streaming response
response_stream = requests.post(f"{API_BASE_URL}chat", json=request_data, stream=True)
response_stream.raise_for_status()
response_obj = itertools.chain([prefix], (chunk.decode() for chunk in response_stream))
response = st.write_stream(response_obj)
else:
# Non-streaming response
api_response = requests.post(f"{API_BASE_URL}chat", json=request_data)
api_response.raise_for_status()
result = api_response.json()
if "warnings" in result:
warnings = result["warnings"]
response = result.get("response", "")
if normalise_response:
response = postprocess_voice_transcription(response)
response = prefix + response
st.write(response)
except requests.exceptions.RequestException as e:
error_msg = re.sub("[a-zA-Z0-9_\-.]+\.com", "