import spaces
import uuid
import warnings
import traceback
import numpy as np
from pathlib import Path
from typing import Optional
from collections import Counter
import gradio as gr
import torch
import torchaudio
import soundfile as sf
import matplotlib.pyplot as plt
from NatureLM.config import Config
from NatureLM.models.NatureLM import NatureLM
from NatureLM.infer import Pipeline
from data_store import upload_data
warnings.filterwarnings("ignore")
SAMPLE_RATE = 16000 # Default sample rate for NatureLM-audio
DEVICE: str = "cuda" if torch.cuda.is_available() else "cpu"
MIN_AUDIO_DURATION: float = 0.5 # seconds
MAX_HISTORY_TURNS = (
3 # Maximum number of conversation turns to include in context (user + assistant pairs)
)
# Load model at startup if CUDA is available
print(f"Device: {DEVICE}")
model = NatureLM.from_pretrained("EarthSpeciesProject/NatureLM-audio")
model = model.eval().to(DEVICE)
model = Pipeline(model)
def check_audio_duration_greater(audio_path: str) -> bool:
"""Check the duration of the audio file."""
info = sf.info(audio_path)
duration = info.duration # info.num_frames / info.sample_rate
if not duration >= MIN_AUDIO_DURATION:
raise gr.Error(f"Audio duration must be at least {MIN_AUDIO_DURATION} seconds.")
def get_spectrogram(audio: torch.Tensor) -> plt.Figure:
"""Generate a spectrogram from the audio tensor."""
spectrogram = torchaudio.transforms.Spectrogram(n_fft=1024)(audio)
spectrogram = spectrogram.numpy()[0].squeeze()
# Convert to matplotlib figure with imshow
fig, ax = plt.subplots(figsize=(13, 5))
ax.imshow(np.log(spectrogram + 1e-4), aspect="auto", origin="lower", cmap="viridis")
ax.set_title("Spectrogram")
ax.set_xlabel("Time")
# Set x ticks to reflect 0 to audio duration seconds
if audio.dim() > 1:
duration = audio.size(1) / SAMPLE_RATE
else:
duration = audio.size(0) / SAMPLE_RATE
ax.set_xticks([0, spectrogram.shape[1]])
ax.set_xticklabels(["0s", f"{duration:.2f}s"])
ax.set_ylabel("Frequency")
# Set y ticks to reflect 0 to nyquist frequency (sample_rate/2)
nyquist_freq = SAMPLE_RATE / 2
ax.set_yticks(
[
0,
spectrogram.shape[0] // 4,
spectrogram.shape[0] // 2,
3 * spectrogram.shape[0] // 4,
spectrogram.shape[0] - 1,
]
)
ax.set_yticklabels(
[
"0 Hz",
f"{nyquist_freq / 4:.0f} Hz",
f"{nyquist_freq / 2:.0f} Hz",
f"{3 * nyquist_freq / 4:.0f} Hz",
f"{nyquist_freq:.0f} Hz",
]
)
fig.tight_layout()
return fig
def take_majority_vote(results: list[list[dict]]) -> list[str]:
"""For each audio file, take the majority vote of the labels across all windows"""
outputs = []
for result in results:
predictions = [window["prediction"] for window in result]
if not predictions:
continue
# Count occurrences of each label
counts = Counter(predictions)
# Find the most common label
most_common_label, _ = counts.most_common(1)[0]
outputs.append(most_common_label)
return outputs
@spaces.GPU
def prompt_lm(
audios: list[str],
queries: list[str] | str,
window_length_seconds: float = 10.0,
hop_length_seconds: float = 10.0,
) -> list[str]:
"""Generate response using the model
Args:
audios (list[str]): List of audio file paths
queries (list[str] | str): Query or list of queries to process
window_length_seconds (float): Length of the window for processing audio
hop_length_seconds (float): Hop length for processing audio
Returns:
list[str]: List of generated responses for each audio-query pair
"""
if model is None:
return "β Model not loaded. Please check the model configuration."
with torch.amp.autocast(device_type="cuda", dtype=torch.float16):
results: list[list[dict]] = model(
audios,
queries,
window_length_seconds=window_length_seconds,
hop_length_seconds=hop_length_seconds,
input_sample_rate=None,
)
return results
def make_spectrogram_figure(audio_input: str) -> list[dict]:
# Load audio with torchaudio and compute spectrogram
if not audio_input:
# Return an empty figure if no audio input is provided
return get_spectrogram(torch.zeros(1, SAMPLE_RATE))
# Check if file exists and is accessible
try:
if not Path(audio_input).exists():
print(f"Audio file does not exist: {audio_input}")
return get_spectrogram(torch.zeros(1, SAMPLE_RATE))
if not Path(audio_input).is_file():
print(f"Path is not a valid file: {audio_input}")
return get_spectrogram(torch.zeros(1, SAMPLE_RATE))
audio_tensor, sample_rate = torchaudio.load(audio_input)
spectrogram_fig = get_spectrogram(audio_tensor)
return spectrogram_fig
except Exception as e:
print(f"Error loading audio file {audio_input}: {e}")
# Return an empty spectrogram on error
return get_spectrogram(torch.zeros(1, SAMPLE_RATE))
def add_user_query(chatbot_history: list[dict], chat_input: str) -> list[dict]:
"""Add user message to chat and get model response"""
# Validate input
if not chat_input.strip():
return chatbot_history
chatbot_history.append({"role": "user", "content": chat_input.strip()})
return chatbot_history
def send_data_to_hub(chatbot_history: list[dict], audio: str, session_id: str):
"""Upload data to hub"""
if not chatbot_history or len(chatbot_history) < 2:
return
user_text = chatbot_history[-2]["content"]
model_response = chatbot_history[-1]["content"]
upload_data(audio, user_text, model_response, session_id)
def get_response(chatbot_history: list[dict], audio_input: str) -> list[dict]:
"""Generate response from the model based on user input and audio file with conversation history"""
try:
# Warn if conversation is getting long
num_turns = len(chatbot_history)
if num_turns > MAX_HISTORY_TURNS * 2: # Each turn = user + assistant message
gr.Warning(
"β οΈ Long conversations may affect response quality. Consider starting a new conversation with the Clear button."
)
# Build conversation context from history
conversation_context = []
for message in chatbot_history:
if message["role"] == "user":
conversation_context.append(f"User: {message['content']}")
elif message["role"] == "assistant":
conversation_context.append(f"Assistant: {message['content']}")
# Get the last user message
last_user_message = ""
for message in reversed(chatbot_history):
if message["role"] == "user":
last_user_message = message["content"]
break
# Format the full prompt with conversation history
if len(conversation_context) > 2: # More than just the current query
# Include previous turns (limit to last MAX_HISTORY_TURNS exchanges)
# recent_context = conversation_context[
# -(MAX_HISTORY_TURNS + 1) : -1
# ] # Exclude current message
recent_context = conversation_context
full_prompt = (
"Previous conversation:\n"
+ "\n".join(recent_context)
+ "\n\nCurrent question: "
+ last_user_message
)
else:
full_prompt = last_user_message
print("\nFull prompt with history:", full_prompt)
response = prompt_lm(
audios=[audio_input],
queries=[full_prompt.strip()],
window_length_seconds=100_000,
hop_length_seconds=100_000,
)
# get first item
if isinstance(response, list) and len(response) > 0:
response = response[0][0]["prediction"]
print("Model response:", response)
else:
response = "No response generated."
except Exception as e:
print(f"Error generating response: {e}")
traceback.print_exc()
response = "Error generating response. Please try again."
# Add model response to chat history
chatbot_history.append({"role": "assistant", "content": response})
return chatbot_history
def main(
assets_dir: Path,
):
# Check if assets directory exists, if not create a placeholder
if not assets_dir.exists():
print(f"Warning: Assets directory {assets_dir} does not exist")
assets_dir.mkdir(exist_ok=True)
# Create placeholder audio files if they don't exist
laz_audio = assets_dir / "Lazuli_Bunting_yell-YELLLAZB20160625SM303143.mp3"
frog_audio = assets_dir / "nri-GreenTreeFrogEvergladesNP.mp3"
robin_audio = assets_dir / "yell-YELLAMRO20160506SM3.mp3"
whale_audio = assets_dir / "Humpback Whale - Megaptera novaeangliae.wav"
crow_audio = assets_dir / "American Crow - Corvus brachyrhynchos.mp3"
examples = {
"Identifying Focal Species (Lazuli Bunting)": [
str(laz_audio),
"What is the common name for the focal species in the audio?",
],
"Caption the audio (Green Tree Frog)": [
str(frog_audio),
"Caption the audio, using the common name for any animal species.",
],
"Caption the audio (American Robin)": [
str(robin_audio),
"Caption the audio, using the scientific name for any animal species.",
],
"Identifying Focal Species (Megaptera novaeangliae)": [
str(whale_audio),
"What is the scientific name for the focal species in the audio?",
],
"Speaker Count (American Crow)": [
str(crow_audio),
"How many individuals are vocalizing in this audio?",
],
"Caption the audio (Humpback Whale)": [str(whale_audio), "Caption the audio."],
}
gr.set_static_paths(paths=[Path.cwd().absolute() / "assets"])
with gr.Blocks(
title="NatureLM-audio",
theme=gr.themes.Base(primary_hue="blue", font=[gr.themes.GoogleFont("Noto Sans")]),
) as app:
with gr.Row():
gr.HTML("""
NatureLM-audioBETA
""")
with gr.Tabs():
with gr.Tab("Analyze Audio"):
session_id = gr.State(str(uuid.uuid4()))
# uploaded_audio = gr.State()
# Status indicator
# status_text = gr.Textbox(
# value=model_manager.get_status(),
# label="Model Status",
# interactive=False,
# visible=True,
# )
with gr.Column(visible=True) as onboarding_message:
gr.HTML(
"""
π
Welcome to NatureLM-audio!
Upload your first audio file below or select a pre-loaded example below.
""",
padding=False,
)
with gr.Column(visible=True) as upload_section:
audio_input = gr.Audio(
type="filepath",
container=True,
interactive=True,
sources=["upload"],
)
# check that audio duration is greater than MIN_AUDIO_DURATION
# raise
audio_input.change(
fn=check_audio_duration_greater,
inputs=[audio_input],
outputs=[],
)
with gr.Accordion(
label="Toggle Spectrogram", open=False, visible=False
) as spectrogram:
plotter = gr.Plot(
get_spectrogram(torch.zeros(1, SAMPLE_RATE)),
label="Spectrogram",
visible=False,
elem_id="spectrogram-plot",
)
with gr.Column(visible=False) as tasks:
task_dropdown = gr.Dropdown(
[
"What are the common names for the species in the audio, if any?",
"Caption the audio, using the scientific name for any animal species.",
"Caption the audio, using the common name for any animal species.",
"What is the scientific name for the focal species in the audio?",
"What is the common name for the focal species in the audio?",
"What is the family of the focal species in the audio?",
"What is the genus of the focal species in the audio?",
"What is the taxonomic name of the focal species in the audio?",
"What call types are heard from the focal species in the audio?",
"What is the life stage of the focal species in the audio?",
],
label="Pre-Loaded Tasks",
info="Select a task, or write your own prompt below.",
allow_custom_value=False,
value=None,
)
with gr.Group(visible=False) as chat:
chatbot = gr.Chatbot(
elem_id="chatbot",
height=250,
type="messages",
label="Chat",
render_markdown=False,
group_consecutive_messages=False,
feedback_options=[
"like",
"dislike",
"wrong species",
"incorrect response",
"other",
],
resizeable=True,
)
with gr.Column() as text:
chat_input = gr.Textbox(
placeholder="Type your message and press Enter to send",
type="text",
lines=1,
show_label=False,
submit_btn="Send",
container=True,
autofocus=False,
elem_id="chat-input",
)
with gr.Column() as examples_section:
gr.Examples(
list(examples.values()),
[audio_input, chat_input],
[audio_input, chat_input],
example_labels=list(examples.keys()),
examples_per_page=20,
)
def validate_and_submit(chatbot_history, chat_input):
if not chat_input or not chat_input.strip():
gr.Warning("Please enter a question or message before sending.")
return chatbot_history, chat_input
updated_history = add_user_query(chatbot_history, chat_input)
return updated_history, ""
clear_button = gr.ClearButton(
components=[chatbot, chat_input, audio_input, plotter],
visible=False,
)
# if task_dropdown is selected, set chat_input to that value
def set_query(task):
if task:
return gr.update(value=task)
return gr.update(value="")
task_dropdown.select(
fn=set_query,
inputs=[task_dropdown],
outputs=[chat_input],
)
def start_chat_interface(audio_path):
return (
gr.update(visible=False), # hide onboarding message
gr.update(visible=True), # show upload section
gr.update(visible=True), # show spectrogram
gr.update(visible=True), # show tasks
gr.update(visible=True), # show chat box
gr.update(visible=True), # show plotter
)
# When audio added, set spectrogram
audio_input.change(
fn=start_chat_interface,
inputs=[audio_input],
outputs=[
onboarding_message,
upload_section,
spectrogram,
tasks,
chat,
plotter,
],
).then(
fn=make_spectrogram_figure,
inputs=[audio_input],
outputs=[plotter],
)
# When submit clicked first:
# 1. Validate and add user query to chat history
# 2. Get response from model
# 3. Clear the chat input box
# 4. Show clear button
chat_input.submit(
validate_and_submit,
inputs=[chatbot, chat_input],
outputs=[chatbot, chat_input],
).then(
get_response,
inputs=[chatbot, audio_input],
outputs=[chatbot],
).then(
lambda: gr.update(visible=True), # Show clear button
None,
[clear_button],
).then(
send_data_to_hub,
[chatbot, audio_input, session_id],
None,
)
clear_button.click(lambda: gr.ClearButton(visible=False), None, [clear_button])
with gr.Tab("Sample Library"):
with gr.Row():
with gr.Column():
gr.Markdown("### Download Sample Audio")
gr.Markdown(
"""Feel free to explore these sample audio files. To download, click the button in the top-right corner of each audio file. You can also find a large collection of publicly available animal sounds on
[Xenocanto](https://xeno-canto.org/explore/taxonomy) and [Watkins Marine Mammal Sound Database](https://whoicf2.whoi.edu/science/B/whalesounds/index.cfm)."""
)
samples = [
(
"assets/Lazuli_Bunting_yell-YELLLAZB20160625SM303143.m4a",
"Lazuli Bunting",
),
(
"assets/nri-GreenTreeFrogEvergladesNP.mp3",
"Green Tree Frog",
),
(
"assets/American Crow - Corvus brachyrhynchos.mp3",
"American Crow",
),
(
"assets/Gray Wolf - Canis lupus italicus.m4a",
"Gray Wolf",
),
(
"assets/Humpback Whale - Megaptera novaeangliae.wav",
"Humpback Whale",
),
("assets/Walrus - Odobenus rosmarus.wav", "Walrus"),
]
for row_i in range(0, len(samples), 3):
with gr.Row():
for filepath, label in samples[row_i : row_i + 3]:
with gr.Column():
gr.Audio(
filepath,
label=label,
type="filepath",
show_download_button=True,
)
with gr.Tab("π‘ Help"):
gr.HTML("""
Help us improve the model!
Found an issue or have suggestions? Join us on Discourse to share feedback and questions.
Upload your audio or click on a pre-loaded example. Drag and drop your audio file containing animal vocalizations, or click on an example.
Trim your audio (if needed) by clicking the scissors icon on the bottom right of the audio panel. Try to keep your audio to 10 seconds or less.
View the Spectrogram (optional). You can easily view/hide the spectrogram of your audio for closer analysis.
Select a task or write your own. Select an option from pre-loaded tasks. This will auto-fill the text box with a prompt, so all you have to do is hit Send. Or, type a custom prompt directly into the chat.
Send and Analyze Audio. Press "Send" or type Enter to begin processing your audio. Ask follow-up questions or press "Clear" to start a new conversation.
Tips
Prompting Best Practices
When possible, use scientific or taxonomic names and mention the context if known (geographic area/location, time of day or year, habitat type)
Ask one question at a time, and be specific about what you want to know
β Don't ask: "Analyze this audio and tell me all you know about it."
β Do ask: "What species made this sound?"
Keep prompts more open-ended and avoid asking Yes/No or very targeted questions
β Don't ask: "Is there a bottlenose dolphin vocalizing in the audio? Yes or No."
β Do ask: "What focal species, if any, are heard in the audio?"
Giving the model options to choose works well for broader categories (less so for specific species)
β Don't ask: "Classify the audio into one of the following species: Bottlenose Dolphin, Orca, Great Gray Owl"β
β Do ask: "Classify the audio into one of the following categories: Cetaceans, Aves, or None."
