Différences

Ci-dessous, les différences entre deux révisions de la page.

--- reconnaissance_audio [2024/05/23 15:57] – créée joseph
+++ reconnaissance_audio [2024/05/23 15:58] (Version actuelle) – joseph
@@ Ligne 1: / Ligne 1: @@
 Test:
-import numpy as np, scipy.signal, glob, subprocess, collections, os, pickle, scipy.io.wavfile
+    import numpy as np, scipy.signal, glob, subprocess, collections, os, pickle, scipy.io.wavfile
-spectrogram = lambda x: 20 * np.log10(np.abs(scipy.signal.stft(x, nperseg=1024, noverlap=1)[2]))
+    spectrogram = lambda x: 20 * np.log10(np.abs(scipy.signal.stft(x, nperseg=1024, noverlap=1)[2]))
-A = 0
+    A = 0
-def add_constellations(s, hash_table, song_id):
+    def add_constellations(s, hash_table, song_id):
-    global A
+        global A
-    global_peaks = []
+        global_peaks = []
-    for t in range(s.shape[1]):
+        for t in range(s.shape[1]):
-        peaks, prop = scipy.signal.find_peaks(s[:, t], prominence=10)
+            peaks, prop = scipy.signal.find_peaks(s[:, t], prominence=10)
-        global_peaks += [(t, f) for _, f in sorted(zip(prop["prominences"], peaks), reverse=True)][:4]
+            global_peaks += [(t, f) for _, f in sorted(zip(prop["prominences"], peaks), reverse=True)][:4]
-    for i1, (t1, f1) in enumerate(global_peaks):
+        for i1, (t1, f1) in enumerate(global_peaks):
-        for (t2, f2) in global_peaks[i1 + 1:i1 + 20]:
+            for (t2, f2) in global_peaks[i1 + 1:i1 + 20]:
-            delta_t = t2 - t1
+                delta_t = t2 - t1
-            if delta_t <= 1 or delta_t >= 4:
+                if delta_t <= 1 or delta_t >= 4:
-                continue
+                    continue
-            h = f1 + (f2 << 8) + (delta_t << 16)
+                h = f1 + (f2 << 8) + (delta_t << 16)
-            hash_table[h].append((t1, song_id) if song_id is not None else t1)
+                hash_table[h].append((t1, song_id) if song_id is not None else t1)
-            A += 1
+                A += 1
-def create_library(path, db_filename, FFMPEG_PATH=r"D:\Documents\software\portable\youtube-dl\ffmpeg.exe"):
+    def create_library(path, db_filename, FFMPEG_PATH=r"D:\Documents\software\portable\youtube-dl\ffmpeg.exe"):
-    hash_table, songs = collections.defaultdict(list), []
+        hash_table, songs = collections.defaultdict(list), []
-    for i, f in enumerate(glob.glob(path)):
+        for i, f in enumerate(glob.glob(path)):
-        print(f"adding to library: {f}, current size: {len(hash_table)=} {A/(i+1)=:,}")
+            print(f"adding to library: {f}, current size: {len(hash_table)=} {A/(i+1)=:,}")
-        songs.append(os.path.basename(f))
+            songs.append(os.path.basename(f))
-        p = subprocess.Popen([FFMPEG_PATH, '-i', f, '-f', 's16le', '-acodec', 'pcm_s16le', '-ar', '44100', '-ac', '1', "-hide_banner", "-loglevel", "fatal", "-"], stdout=subprocess.PIPE, bufsize=10**8)
+            p = subprocess.Popen([FFMPEG_PATH, '-i', f, '-f', 's16le', '-acodec', 'pcm_s16le', '-ar', '44100', '-ac', '1', "-hide_banner", "-loglevel", "fatal", "-"], stdout=subprocess.PIPE, bufsize=10**8)
-        x = np.frombuffer(p.communicate()[0], dtype="int16")
+            x = np.frombuffer(p.communicate()[0], dtype="int16")
-        s = spectrogram(x)[:256]
+            s = spectrogram(x)[:256]
-        add_constellations(s, hash_table, i)
+            add_constellations(s, hash_table, i)
-        if i == 200:
+            if i == 200:
-            break
+                break
-    with open(db_filename, "wb") as g:
+        with open(db_filename, "wb") as g:
-        pickle.dump({"hash_table": hash_table, "songs": songs}, g)
+            pickle.dump({"hash_table": hash_table, "songs": songs}, g)
-    print("create_library: finished.")
+        print("create_library: finished.")
-def load_library(db_filename="db.db"):
+    def load_library(db_filename="db.db"):
-    return pickle.load(open(db_filename, "rb"))
+        return pickle.load(open(db_filename, "rb"))
-def recognize(f, db):
+    def recognize(f, db):
-    sr, x = scipy.io.wavfile.read(f)
+        sr, x = scipy.io.wavfile.read(f)
-    s = spectrogram(x)
+        s = spectrogram(x)
-    recording_hash_table = collections.defaultdict(list)
+        recording_hash_table = collections.defaultdict(list)
-    matches_per_song = collections.defaultdict(list)
+        matches_per_song = collections.defaultdict(list)
-    scores = collections.defaultdict(int)
+        scores = collections.defaultdict(int)
-    add_constellations(s, recording_hash_table, None)
+        add_constellations(s, recording_hash_table, None)
-    for h, T in recording_hash_table.items():
+        for h, T in recording_hash_table.items():
-        for t1 in T:
+            for t1 in T:
-            for (t0, song_id) in db["hash_table"][h]:
+                for (t0, song_id) in db["hash_table"][h]:
-                matches_per_song[song_id].append((h, t1, t0))
+                    matches_per_song[song_id].append((h, t1, t0))
-    for song_id, matches in matches_per_song.items():
+        for song_id, matches in matches_per_song.items():
-        song_scores_by_offset = collections.defaultdict(int)
+            song_scores_by_offset = collections.defaultdict(int)
-        for h, t1, t0 in matches:
+            for h, t1, t0 in matches:
-            song_scores_by_offset[t0 - t1] += 1
+                song_scores_by_offset[t0 - t1] += 1
-        scores[song_id] = max(song_scores_by_offset.items(), key=lambda x: x[1])
+            scores[song_id] = max(song_scores_by_offset.items(), key=lambda x: x[1])
-    scores = sorted(scores.items(), key=lambda x: x[1][1], reverse=True)
+        scores = sorted(scores.items(), key=lambda x: x[1][1], reverse=True)
-    print(f, db["songs"][scores[0][0]], scores)
+        print(f, db["songs"][scores[0][0]], scores)
-# create_library(r"D:\Documents\mp3\_misc\*.mp3", "misc.db")
+    # create_library(r"D:\Documents\mp3\_misc\*.mp3", "misc.db")
-db = load_library("misc.db")
+    db = load_library("misc.db")
-recognize("test6_ragazzo.wav", db)
+    recognize("test6_ragazzo.wav", db)