initial commit

chord_harp/ChordEstimation

@@ -0,0 +1 @@
+Subproject commit bdd05b2236130f01ca2875a30297f3f368fbf470

chord_harp/ChordRecognitionMIDITrainedExtractor

@@ -0,0 +1 @@
+Subproject commit 19b81f694dae7782111b1be0e39e66f42fa1615f

chord_harp/Dockerfile

@@ -0,0 +1,34 @@
+FROM python:3.8-slim
+
+WORKDIR /app
+
+# Install system dependencies
+COPY packages.txt /app/packages.txt
+RUN apt-get update && \
+    xargs apt-get install -y --no-install-recommends < /app/packages.txt && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install Python dependencies
+COPY requirements.txt /app/requirements.txt
+ENV PIP_NO_BUILD_ISOLATION=1
+RUN pip install --no-cache-dir -U pip wheel Cython
+RUN pip install --no-cache-dir setuptools==59.8.0
+RUN pip install --no-cache-dir numpy==1.23.5
+RUN pip install --no-cache-dir -r /app/requirements.txt
+RUN pip install --no-cache-dir --no-build-isolation madmom
+
+# Apply madmom patch
+COPY patch_madmom.py /app/patch_madmom.py
+RUN python /app/patch_madmom.py
+RUN python -c "import madmom; print('madmom import OK')"
+
+# Copy the rest of the app
+COPY . /app
+
+# Set working directory to chord recognition repo so model files are found
+WORKDIR /app/ChordRecognitionMIDITrainedExtractor
+
+ENV PORT=7860
+EXPOSE 7860
+
+CMD ["python", "/app/app.py"]

chord_harp/README.md

@@ -0,0 +1,21 @@
+---
+title: Automatic Chord Recognition
+emoji: 🎵
+colorFrom: blue
+colorTo: purple
+sdk: docker
+app_port: 7860
+---
+
+# Automatic Chord Recognition (PyHARP)
+
+This is a [PyHARP](https://github.com/TEAMuP-dev/pyharp) app that estimates chord progressions from audio using the model proposed in:
+
+> Wu, Y., & Li, W. (2019). Automatic Audio Chord Recognition With MIDI-Trained Deep Feature and BLSTM-CRF Sequence Decoding Model. *IEEE/ACM Transactions on Audio, Speech, and Language Processing*, 27(2), 355–366.
+
+## How it works
+1. Upload an audio file in HARP
+2. The model extracts a Harmonic CQT spectrogram
+3. A CNN extracts pitch class features (trained on MIDI data)
+4. A BLSTM-CRF decodes the chord label sequence
+5. Time-stamped chord labels are returned and displayed in HARP

chord_harp/app.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+import sys
+import os
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "ChordRecognitionMIDITrainedExtractor"))
+
+import numpy as np
+import networks as N
+from librosa.core import cqt, load, note_to_hz
+import const as C
+import utils as U
+
+from pyharp import ModelCard, build_endpoint, LabelList, AudioLabel, OutputLabel
+import gradio as gr
+
+# ── Load models once at startup ──────────────────────────────────────────────
+cnn_feat_extractor = N.FullCNNFeatExtractor()
+cnn_feat_extractor.load(C.DEFAULT_CONVNETFILE)
+
+decoder = N.NBLSTMCRF()
+decoder.load("nblstm_crf.model")
+
+# ── Model card ───────────────────────────────────────────────────────────────
+model_card = ModelCard(
+    name="Automatic Chord Recognition",
+    description="Estimates chord progressions from audio using a CNN feature extractor trained on MIDI data and a BLSTM-CRF sequence decoder. Outputs time-stamped chord labels.",
+    author="Wu & Li (2019), wrapped by PyHARP",
+    tags=["chord recognition", "harmony", "MIR"],
+)
+
+# ── Processing function ───────────────────────────────────────────────────────
+def process_fn(input_audio_path: str) -> LabelList:
+    # Load audio
+    y, sr = load(input_audio_path, sr=C.SR)
+
+    # Extract Harmonic-CQT
+    fmin = note_to_hz("C1")
+    hcqt = np.stack([
+        np.abs(cqt(
+            y, sr=C.SR, hop_length=C.H, n_bins=C.BIN_CNT,
+            bins_per_octave=C.OCT_BIN, fmin=fmin * (h + 1),
+            filter_scale=2, tuning=None
+        )).T.astype(np.float32)
+        for h in range(C.CQT_H)
+    ])
+
+    # Extract deep feature
+    feat = cnn_feat_extractor.GetFeature(U.PreprocessSpec(hcqt)).data
+
+    # Decode chord label sequence
+    labels = decoder.argmax(feat)
+
+    # Build LabelList for HARP
+    output_labels = LabelList()
+    cur_label = labels[0]
+    st = 0
+
+    for i in range(labels.size):
+        if labels[i] != cur_label:
+            ed = i
+            feat_seg = feat[st:ed, :]
+            chord_sign = U.voc.ChordSignature7thbass(cur_label, feat_seg, sevenths=True, inv=True)
+            start_sec = float(st * C.H) / C.SR
+            end_sec = float(ed * C.H) / C.SR
+
+            if chord_sign != "N":
+                output_labels.labels.append(
+                    AudioLabel(
+                        t=start_sec,
+                        label=chord_sign,
+                        duration=end_sec - start_sec,
+                        description=f"Chord: {chord_sign} ({start_sec:.2f}s - {end_sec:.2f}s)",
+                    )
+                )
+            cur_label = labels[i]
+            st = i
+
+    # Handle last segment
+    feat_seg = feat[st:labels.size, :]
+    chord_sign = U.voc.ChordSignature7thbass(cur_label, feat_seg)
+    start_sec = float(st * C.H) / C.SR
+    end_sec = float(labels.size * C.H) / C.SR
+    if chord_sign != "N":
+        output_labels.labels.append(
+            AudioLabel(
+                t=start_sec,
+                label=chord_sign,
+                duration=end_sec - start_sec,
+                description=f"Chord: {chord_sign} ({start_sec:.2f}s - {end_sec:.2f}s)",
+            )
+        )
+
+    return output_labels
+
+
+# ── Gradio endpoint ───────────────────────────────────────────────────────────
+with gr.Blocks() as demo:
+    input_components = [
+        gr.Audio(type="filepath", label="Input Audio").harp_required(True),
+    ]
+
+    output_components = [
+        gr.JSON(label="Output Labels"),
+    ]
+
+    app = build_endpoint(
+        model_card=model_card,
+        input_components=input_components,
+        output_components=output_components,
+        process_fn=process_fn,
+    )
+
+demo.queue().launch(share=True, show_error=False, pwa=True)

chord_harp/packages.txt

@@ -0,0 +1,5 @@
+gcc
+g++
+libsndfile1
+ffmpeg
+git

chord_harp/patch_madmom.py

@@ -0,0 +1,33 @@
+#!/usr/bin/env python3
+"""
+Patch madmom to fix deprecated sklearn and other compatibility issues.
+"""
+import os
+import site
+
+def patch_file(filepath, old, new):
+    if not os.path.exists(filepath):
+        print(f"Skipping (not found): {filepath}")
+        return
+    with open(filepath, "r") as f:
+        content = f.read()
+    if old in content:
+        content = content.replace(old, new)
+        with open(filepath, "w") as f:
+            f.write(content)
+        print(f"Patched: {filepath}")
+    else:
+        print(f"No patch needed: {filepath}")
+
+# Find madmom install location
+site_packages = site.getsitepackages()[0]
+madmom_base = os.path.join(site_packages, "madmom")
+
+# Fix sklearn.externals.joblib import
+patch_file(
+    os.path.join(madmom_base, "ml", "hmm.py"),
+    "from sklearn.externals import joblib",
+    "import joblib"
+)
+
+print("Madmom patching complete.")

chord_harp/requirements.txt

@@ -0,0 +1,7 @@
+gradio==5.28.0
+git+https://github.com/TEAMuP-dev/pyharp.git@v0.3.0
+chainer==7.8.1
+librosa
+scipy
+mir_eval
+joblib