add option instruction. add evaluation results.

aikiriao · Aug 29, 2024 · 3bd1629 · 3bd1629
1 parent b5625df
commit 3bd1629
Show file tree

Hide file tree

Showing 38 changed files with 161 additions and 2 deletions.
diff --git a/README.md b/README.md
@@ -32,18 +32,48 @@ cmake --build build
 ./srla -e INPUT.wav OUTPUT.srl
 ```
 
-you can change compression mode by `-m` option.
+#### Mode `-m`
+
+You can change compression mode by `-m` option.
 Following example encoding in maximum compression (but slow) option.
 
 ```bash
-./srla -e -m 5 INPUT.wav OUTPUT.srl
+./srla -e -m 16 INPUT.wav OUTPUT.srl
+```
+#### Max block size `-B`
+
+You can change maximum block size by `-B` option.
+Following example encoding with maximum block size be 4096 sample.
+
+```bash
+./srla -e -B 4096 INPUT.wav OUTPUT.srl
+```
+
+#### Number of divisions in block `-V`
+
+You can change number of division by `-V` option. Number of division specifies search depth of optimal block division.
+Following example encoding with number of divisions in block to be $2^{2} = 4$.
+
+```bash
+./srla -e -V 2 INPUT.wav OUTPUT.srl
 ```
 
 ### Decode
 
 ```bash
 ./srla -d INPUT.srl OUTPUT.wav
 ```
+## Performance
+
+We use [RWC music dataset](https://staff.aist.go.jp/m.goto/RWC-MDB/) for comparison.
+
+### Decoding speed vs compression ratio
+
+![Decoding speed vs compression rate](./evaluation/decodespeed_vs_compressionrate_8192_total.png)
+
+### Encoding speed vs compression ratio
+
+![Encoding speed vs compression rate](./evaluation/encodespeed_vs_compressionrate_8192_total.png)
 
 ## License
 

diff --git a/evaluation/decodespeed_vs_compressionrate_16384_classic.png b/evaluation/decodespeed_vs_compressionrate_16384_classic.png
diff --git a/evaluation/decodespeed_vs_compressionrate_16384_genre.png b/evaluation/decodespeed_vs_compressionrate_16384_genre.png
diff --git a/evaluation/decodespeed_vs_compressionrate_16384_jazz.png b/evaluation/decodespeed_vs_compressionrate_16384_jazz.png
diff --git a/evaluation/decodespeed_vs_compressionrate_16384_popular.png b/evaluation/decodespeed_vs_compressionrate_16384_popular.png
diff --git a/evaluation/decodespeed_vs_compressionrate_16384_right.png b/evaluation/decodespeed_vs_compressionrate_16384_right.png
diff --git a/evaluation/decodespeed_vs_compressionrate_16384_total.png b/evaluation/decodespeed_vs_compressionrate_16384_total.png
diff --git a/evaluation/decodespeed_vs_compressionrate_4096_classic.png b/evaluation/decodespeed_vs_compressionrate_4096_classic.png
diff --git a/evaluation/decodespeed_vs_compressionrate_4096_genre.png b/evaluation/decodespeed_vs_compressionrate_4096_genre.png
diff --git a/evaluation/decodespeed_vs_compressionrate_4096_jazz.png b/evaluation/decodespeed_vs_compressionrate_4096_jazz.png
diff --git a/evaluation/decodespeed_vs_compressionrate_4096_popular.png b/evaluation/decodespeed_vs_compressionrate_4096_popular.png
diff --git a/evaluation/decodespeed_vs_compressionrate_4096_right.png b/evaluation/decodespeed_vs_compressionrate_4096_right.png
diff --git a/evaluation/decodespeed_vs_compressionrate_4096_total.png b/evaluation/decodespeed_vs_compressionrate_4096_total.png
diff --git a/evaluation/decodespeed_vs_compressionrate_8192_classic.png b/evaluation/decodespeed_vs_compressionrate_8192_classic.png
diff --git a/evaluation/decodespeed_vs_compressionrate_8192_genre.png b/evaluation/decodespeed_vs_compressionrate_8192_genre.png
diff --git a/evaluation/decodespeed_vs_compressionrate_8192_jazz.png b/evaluation/decodespeed_vs_compressionrate_8192_jazz.png
diff --git a/evaluation/decodespeed_vs_compressionrate_8192_popular.png b/evaluation/decodespeed_vs_compressionrate_8192_popular.png
diff --git a/evaluation/decodespeed_vs_compressionrate_8192_right.png b/evaluation/decodespeed_vs_compressionrate_8192_right.png
diff --git a/evaluation/decodespeed_vs_compressionrate_8192_total.png b/evaluation/decodespeed_vs_compressionrate_8192_total.png
diff --git a/evaluation/encodespeed_vs_compressionrate_16384_classic.png b/evaluation/encodespeed_vs_compressionrate_16384_classic.png
diff --git a/evaluation/encodespeed_vs_compressionrate_16384_genre.png b/evaluation/encodespeed_vs_compressionrate_16384_genre.png
diff --git a/evaluation/encodespeed_vs_compressionrate_16384_jazz.png b/evaluation/encodespeed_vs_compressionrate_16384_jazz.png
diff --git a/evaluation/encodespeed_vs_compressionrate_16384_popular.png b/evaluation/encodespeed_vs_compressionrate_16384_popular.png
diff --git a/evaluation/encodespeed_vs_compressionrate_16384_right.png b/evaluation/encodespeed_vs_compressionrate_16384_right.png
diff --git a/evaluation/encodespeed_vs_compressionrate_16384_total.png b/evaluation/encodespeed_vs_compressionrate_16384_total.png
diff --git a/evaluation/encodespeed_vs_compressionrate_4096_classic.png b/evaluation/encodespeed_vs_compressionrate_4096_classic.png
diff --git a/evaluation/encodespeed_vs_compressionrate_4096_genre.png b/evaluation/encodespeed_vs_compressionrate_4096_genre.png
diff --git a/evaluation/encodespeed_vs_compressionrate_4096_jazz.png b/evaluation/encodespeed_vs_compressionrate_4096_jazz.png
diff --git a/evaluation/encodespeed_vs_compressionrate_4096_popular.png b/evaluation/encodespeed_vs_compressionrate_4096_popular.png
diff --git a/evaluation/encodespeed_vs_compressionrate_4096_right.png b/evaluation/encodespeed_vs_compressionrate_4096_right.png
diff --git a/evaluation/encodespeed_vs_compressionrate_4096_total.png b/evaluation/encodespeed_vs_compressionrate_4096_total.png
diff --git a/evaluation/encodespeed_vs_compressionrate_8192_classic.png b/evaluation/encodespeed_vs_compressionrate_8192_classic.png
diff --git a/evaluation/encodespeed_vs_compressionrate_8192_genre.png b/evaluation/encodespeed_vs_compressionrate_8192_genre.png
diff --git a/evaluation/encodespeed_vs_compressionrate_8192_jazz.png b/evaluation/encodespeed_vs_compressionrate_8192_jazz.png
diff --git a/evaluation/encodespeed_vs_compressionrate_8192_popular.png b/evaluation/encodespeed_vs_compressionrate_8192_popular.png
diff --git a/evaluation/encodespeed_vs_compressionrate_8192_right.png b/evaluation/encodespeed_vs_compressionrate_8192_right.png
diff --git a/evaluation/encodespeed_vs_compressionrate_8192_total.png b/evaluation/encodespeed_vs_compressionrate_8192_total.png
diff --git a/evaluation/plot_codec_performance_graph.py b/evaluation/plot_codec_performance_graph.py
@@ -0,0 +1,129 @@
+" コーデック評価グラフの作成 "
+import pandas as pd
+import matplotlib
+import matplotlib.pyplot as plt
+from adjustText import adjust_text
+
+# type3 font 回避(tex使用)
+matplotlib.rcParams['text.usetex'] = True
+matplotlib.rcParams['text.latex.preamble'] = '\\usepackage{sfmath}'
+# フォントサイズ一括設定
+matplotlib.rcParams["font.size"] = 12
+matplotlib.rcParams['pgf.texsystem'] = 'lualatex'
+
+OTHER_CODEC_LABEL_PREFIXES = ['FLAC', 'WavPack', 'TTA', 'Monkey\'s Audio', 'MPEG4-ALS', 'TAK']
+COLORLIST = ['crimson', 'g', 'b', 'c', 'm', 'k', 'red', 'orange']
+CATEGORIES = ['classic', 'genre', 'jazz', 'popular', 'right', 'total']
+
+AVOID_LABEL_LIST = ['FLAC -0', 'WavPack -x4', 'WavPack -h -x4', 'WavPack -hh -x4', 'Monkey\'s Audio -c4000', 'MPEG4-ALS -b', 'MPEG4-ALS -7', 'TAK -p0e', 'TAK -p0', 'TAK -p1e', 'TAK -p1', 'TAK -p2e', 'TAK -p2', 'TAK -p3e', 'TAK -p3','TAK -p4e', 'TAK -p4']
+
+def _is_avoid_label(label):
+    for content in AVOID_LABEL_LIST:
+        if label == content:
+            return True
+    return False
+
+if __name__ == "__main__":
+    other_codecs_df = pd.read_csv('codec_comparison_summery_other_codecs.csv', index_col=0)
+    srla_codecs_df = pd.read_csv('codec_comparison_summery.csv', index_col=0)
+    avx2_srla_codecs_df = pd.read_csv('codec_comparison_summery_avx2.csv', index_col=0)
+
+    for block_size in [4096, 8192, 16384]:
+        # デコード速度 v.s. 圧縮率グラフ
+        for category in CATEGORIES:
+            texts = []
+            plt.cla()
+            # 他コーデック
+            for inx, cprefix in enumerate(OTHER_CODEC_LABEL_PREFIXES):
+                line = [[], []]
+                for label in other_codecs_df.keys():
+                    if _is_avoid_label(label) is True:
+                        continue
+                    if label.startswith(cprefix):
+                        decode_time = other_codecs_df.at[f'{category} mean decode time', label]
+                        compress_rate = other_codecs_df.at[f'{category} mean compression rate', label]
+                        texts.append(plt.text(decode_time, compress_rate, label[len(cprefix):], size=10))
+                        line[0].append(decode_time)
+                        line[1].append(compress_rate)
+                plt.plot(line[0], line[1], color=COLORLIST[inx], label=cprefix, marker='o')
+            # SRLA
+            for div_index, div in enumerate([0, 2]):
+                line = [[], []]
+                for label in srla_codecs_df.keys():
+                    if label.startswith('SRLA'):
+                        if not str(block_size) in label or not f'-V {div}' in label:
+                            continue
+                        option_prefix = label[len('SRLA'):label.index('V') - 1]
+                        decode_time = srla_codecs_df.at[f'{category} mean decode time', label]
+                        compress_rate = srla_codecs_df.at[f'{category} mean compression rate', label]
+                        texts.append(plt.text(decode_time, compress_rate, option_prefix, size=10))
+                        line[0].append(decode_time)
+                        line[1].append(compress_rate)
+                plt.plot(line[0], line[1], color=COLORLIST[len(OTHER_CODEC_LABEL_PREFIXES) + div_index], label=f'SRLA V={div}', marker='^')
+            # AVX2 SRLA
+            for div_index, div in enumerate([0, 2]):
+                line = [[], []]
+                for label in avx2_srla_codecs_df.keys():
+                    if label.startswith('SRLA'):
+                        if not str(block_size) in label or not f'-V {div}' in label:
+                            continue
+                        option_prefix = label[len('SRLA'):label.index('V') - 1]
+                        decode_time = avx2_srla_codecs_df.at[f'{category} mean decode time', label]
+                        compress_rate = avx2_srla_codecs_df.at[f'{category} mean compression rate', label]
+                        texts.append(plt.text(decode_time, compress_rate, option_prefix, size=10))
+                        line[0].append(decode_time)
+                        line[1].append(compress_rate)
+                plt.plot(line[0], line[1], color=COLORLIST[len(OTHER_CODEC_LABEL_PREFIXES) + div_index], label=f'AVX2 SRLA V={div}', marker='^', linestyle='--')
+
+            adjust_text(texts)
+            plt.title(f'Decoding speed v.s. compression rate for {category} blocksize:{block_size}')
+            plt.xlabel('Average decoding speed (\%)')
+            plt.ylabel('Average compression rate (\%)')
+            plt.legend(ncols=2)
+            plt.grid()
+            plt.xlim((0.1, 0.42))
+            if category == 'total':
+                plt.ylim(ymin=53.5)
+            plt.tight_layout()
+            plt.savefig(f'decodespeed_vs_compressionrate_{block_size}_{category}.png')
+
+        # エンコード速度 v.s. 圧縮率グラフ
+        for category in CATEGORIES:
+            texts = []
+            plt.cla()
+            # 他コーデック
+            for inx, cprefix in enumerate(OTHER_CODEC_LABEL_PREFIXES):
+                line = [[], []]
+                for label in other_codecs_df.keys():
+                    if _is_avoid_label(label) is True:
+                        continue
+                    if label.startswith(cprefix):
+                        encode_time = other_codecs_df.at[f'{category} mean encode time', label]
+                        compress_rate = other_codecs_df.at[f'{category} mean compression rate', label]
+                        texts.append(plt.text(encode_time, compress_rate, label[len(cprefix):], size=10))
+                        line[0].append(encode_time)
+                        line[1].append(compress_rate)
+                plt.plot(line[0], line[1], color=COLORLIST[inx], label=cprefix, marker='o')
+            # SRLA
+            for div_index, div in enumerate([0, 2]):
+                line = [[], []]
+                for label in srla_codecs_df.keys():
+                    if label.startswith('SRLA'):
+                        if not str(block_size) in label or not f'-V {div}' in label:
+                            continue
+                        option_prefix = label[len('SRLA'):label.index('V') - 1]
+                        encode_time = srla_codecs_df.at[f'{category} mean encode time', label]
+                        compress_rate = srla_codecs_df.at[f'{category} mean compression rate', label]
+                        texts.append(plt.text(encode_time, compress_rate, option_prefix, size=10))
+                        line[0].append(encode_time)
+                        line[1].append(compress_rate)
+                plt.plot(line[0], line[1], color=COLORLIST[len(OTHER_CODEC_LABEL_PREFIXES) + div_index], label=f'SRLA V={div}', marker='^')
+
+            adjust_text(texts)
+            plt.title(f'Encoding speed v.s. compression rate for {category} blocksize:{block_size}')
+            plt.xlabel('Average encoding speed (\%)')
+            plt.ylabel('Average compression rate (\%)')
+            plt.legend(ncols=2)
+            plt.grid()
+            plt.tight_layout()
+            plt.savefig(f'encodespeed_vs_compressionrate_{block_size}_{category}.png')