# DrumScript/audio_processor/onset_detector.py
"""
This module will detect the onset (start) times of drum hits in the audio.
"""
import argparse
import os
import librosa
import numpy as np
from drumscript.audio_processor.tempo_detector import estimate_tempo
from drumscript.notation_generator.constants import HOP_LENGTH, SAMPLE_RATE
[docs]
def detect_onsets(audio_data: np.ndarray, sr: int) -> list[float]:
# Detects the onset (start) times of percussive events in an audio signal.
# This function uses librosa's built-in onset detection algorithms, which
# typically rely on spectral flux or other energy-based methods to identify
# sudden changes in the audio signal characteristic of percussive hits.
# Args:
# audio_data (np.ndarray): The input audio time series.
# sr (int): The sample rate of the audio data.
# Returns:
# list[float]: A list of detected onset times in seconds.
if audio_data.size == 0:
return []
y_percussive = librosa.effects.percussive(y=audio_data)
# --- ENFORCE PHYSICAL DRUMMING LIMITS ---
# Because HOP_LENGTH is 128, the frames are very tiny (~2.9ms).
# We must explicitly tell Librosa to wait at least 50ms before triggering
# a second hit, otherwise it will trigger on cymbal vibrations.
lockout_time_secs = 0.05
# wait_frames = int(lockout_time_secs * (SAMPLE_RATE / HOP_LENGTH))
wait_frames = int(lockout_time_secs * (sr / HOP_LENGTH))
print(f"(HOP_LENGTH: {HOP_LENGTH})")
print(f"(Wait Frames Applied: {wait_frames})")
# Pass the 'wait' and 'delta' constraints directly into the simple wrapper
onset_frames = librosa.onset.onset_detect(
y=y_percussive,
# sr=SAMPLE_RATE,
sr=sr,
hop_length=HOP_LENGTH,
units="frames",
wait=wait_frames, # Stops rapid double-triggering on cymbals
delta=0.05, # Ignores general background noise floor
)
# Commenting out to test single_beat_refinement :)
# print(f'(len_onset_frames:{len(onset_frames)})')
# Convert onset frames to time in seconds
# onset_times = librosa.frames_to_time(
# onset_frames,
# sr=SAMPLE_RATE,
# hop_length=HOP_LENGTH
# )
# print(f'(len_onset_times:{len(onset_times)})')
# return onset_times.tolist()
# Convert onset frames to time in seconds
# onset_times = librosa.frames_to_time(onset_frames, sr=SAMPLE_RATE, hop_length=HOP_LENGTH).tolist()
onset_times = librosa.frames_to_time(onset_frames, sr=sr, hop_length=HOP_LENGTH).tolist()
# --- SINGLE-BEAT REFINEMENT ---
# If the total duration is very short (< 2.0s), it's likely a single hit sample.
# We apply a stricter "De-bounce" to prevent room reflections from triggering events.
duration = len(audio_data) / sr
# if duration < 1.0 and len(onset_times) > 1:
if duration < 2.0 and len(onset_times) > 1:
# We only keep the FIRST onset if others follow too closely (within 150ms)
# This fixes the 'single hit showing multiple events' issue in orchestration folder.
refined_onsets = [onset_times[0]]
for i in range(1, len(onset_times)):
if onset_times[i] - refined_onsets[-1] > 0.150: # 150ms threshold
refined_onsets.append(onset_times[i])
onset_times = refined_onsets
print(f"(len_onset_times:{len(onset_times)})")
return onset_times
# ------- AUTOMATIC TEMPO DETECTION------------------------------------
# REPLACED THE FUNCTION THAT WAS HARDCODED TO DETECT TEMPO FROM ONSETS WITH IMPORTED FCT FROM THE TEMPO_DETECTOR SCRIPT
"""
def calculate_tempo_from_onsets(onset_times: np.ndarray, sr: int) -> float:
#
#Estimates the tempo (BPM) from a list of onset timestamps.
# :param onset_times: Array of onset timestamps in seconds.
#:type onset_times: np.ndarray
#:param sr: The sampling rate.
#:type sr: int
#:return: The estimated tempo in BPM.
#:rtype: float
#
if len(onset_times) < 2:
return 120.0 # Return a default tempo if not enough onsets are found
# FIX: librosa.beat.tempo does NOT accept 'onset_events' without audio/envelope.
# We calculate tempo using Inter-Onset Intervals (IOI).
# 1. Calculate the time difference between consecutive hits
ioi = np.diff(onset_times)
# 2. Filter out extremely short or long gaps (e.g., fast rolls or long pauses)
# 0.2s = 300 BPM, 1.5s = 40 BPM
valid_ioi = ioi[(ioi > 0.2) & (ioi < 1.5)]
if len(valid_ioi) > 0:
# 3. Take the median interval (median is better than average as it ignores outliers)
avg_interval = np.median(valid_ioi)
tempo = 60.0 / avg_interval
else:
tempo = 120.0 # Default if pattern is weird or too sparse
return float(tempo)
"""
if __name__ == "__main__":
from drumscript.audio_processor.audio_loader import load_audio, normalise_audio
from drumscript.audio_processor.tempo_detector import estimate_tempo
# --------------------------------------------------------------------------uncomment during testing
# from datetime import datetime
# print("\n# ------------------------------------------------------------------------------------")
# datetimestamp = datetime.now()
# print(f'\ndate/time: {datetimestamp}')
# --------------------------------------------------------------------------------------------------
print("Running onset_detector.py example with provided filepath...") # FUTURE: Find way to encode this so it prints the file path provided in CLI
try:
# Import necessary modules from package
# Note: You might need 'from DrumScript.audio_processor.audio_loader import ...'
# if running this script directly and 'audio_processor' is not in the Python path.
# However, for 'python -m' style execution, 'from audio_processor.audio_loader import ...' is usually correct.
# Required for CLI argparsing
parser = argparse.ArgumentParser(description="Detect onsets in drum audio.")
parser.add_argument("audio_path", help="Path to the input audio file")
args = parser.parse_args()
# sr = 44100 # Target sample rate for processing
# sr = 44100*1.5 # Target sample rate for processing
sr = SAMPLE_RATE
print(f"sample_rate=sr={sr}") # Print current sample rate applied
# --- Get path to audio (cli or import from sister module)
current_script_dir = os.path.dirname(os.path.abspath(__file__))
print(f"current_script_dir: {current_script_dir}")
# Go up one level from audio_processor/onset_detector.py to the outer DRUMSCRIPT/ folder
project_root = os.path.abspath(os.path.join(current_script_dir, "..", ".."))
print(f"project_root: {project_root}")
audio_path = os.path.abspath(args.audio_path)
print(f"audio_path: {audio_path}")
print(f"Attempting to load: {audio_path}")
# Load and normalise audio
# audio_data, sample_rate = load_audio(audio_path, sr=sr)
audio_data, sample_rate = load_audio(audio_path, sr=SAMPLE_RATE)
normalised_audio = normalise_audio(audio_data)
# Detect onsets
print(f"Detecting onsets in : {audio_path}")
onsets = detect_onsets(normalised_audio, SAMPLE_RATE)
print(f"Detected {len(onsets)} onsets.")
if onsets:
# Print the first few detected onsets for verification
# print("\nFirst 10 detected onsets (seconds):")
# for i, onset_time in enumerate(onsets[:10]):
# if len(onsets) > 10:
# print(f" ...and {len(onsets) - 10} more onsets.")
# Print * (ALL) detected onsets for now
print(f"\n All {len(onsets)} detected onsets (seconds):")
for i, onset_time in enumerate(onsets):
print(f" Onset {i + 1}: {onset_time:.4f}s")
# for i, onset_time in enumerate(onsets):
# print(f" Onset {i+1}: {onsets:.4f}s")
else:
print(f"No onsets detected in audio_path: {audio_path}")
# global_tempo = estimate_tempo(audio_data, SAMPLE_RATE, HOP_LENGTH)
# tempo = estimate_tempo(audio_data, SAMPLE_RATE, HOP_LENGTH)
tempo = estimate_tempo(audio_data, SAMPLE_RATE)
# tempo = estimate_tempo(audio_data, SAMPLE_RATE)/2 # temporary fix
# tempo = estimate_tempo(audio_data, SAMPLE_RATE)/4 # temporary fix
# print(f"Loaded audio: Shape={normalised_audio.shape}, Sample Rate={sample_rate}, Duration={len(normalised_audio)/sample_rate:.2f} seconds,
# Tempo={calculate_tempo_from_onsets(onsets, sr=SAMPLE_RATE):2f}")
duration = len(normalised_audio) / sample_rate
print(
f"Loaded audio: Shape=[{normalised_audio.shape}, Sample Rate={sample_rate} (Hz), "
f"Hop Length={HOP_LENGTH} (Hz), Duration={duration:.2f} seconds, Tempo={tempo:.2f} BPM]"
)
except FileNotFoundError:
print(f"\nERROR: The audio file '{audio_path}' was not found.")
print(f"\nPlease ensure you have provided the correct path to audio file: {audio_path}")
except ImportError as e:
print(f"\nERROR: Required modules/libraries might be missing or imports are incorrect: {e}")
print("Ensure 'soundfile', 'librosa', 'numpy', and DrumScript modules are correctly installed and structured.")
print("For MP3, 'ffmpeg' must also be installed on system and accessible in PATH.")
except Exception as e:
print(f"\nAn unexpected error occurred during the example execution: {e}")
import traceback
traceback.print_exc() # Print full traceback for debugging
print("\nonset_detector.py example finished.")
# --------------------------------------------------------------------------uncomment during testing
# from datetime import datetime
# print("\n# ------------------------------------------------------------------------------------")
# datetimestamp = datetime.now()
# print(f'\ndate/time: {datetimestamp}')
# --------------------------------------------------------------------------------------------------
DrumScript