Spaces:
Running
Running
srtの分割を1分にし、configなどを整え、READMEを書いた
Browse files- README.md +135 -0
- src/config.py +8 -0
- src/episode.py +8 -13
- src/store.py +4 -8
README.md
CHANGED
@@ -3,3 +3,138 @@
|
|
3 |
Podcast terapyon channelを検索する仕組み
|
4 |
|
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
Podcast terapyon channelを検索する仕組み
|
4 |
|
5 |
|
6 |
+
## 使い方
|
7 |
+
|
8 |
+
|
9 |
+
### タイトルリスト
|
10 |
+
|
11 |
+
- 以下のファイルを`store` フォルダに置く
|
12 |
+
- `title-list-202301-202501.parquet`
|
13 |
+
- 以下のカラムを持つ
|
14 |
+
- id: int
|
15 |
+
- date: str (2023-01-09)
|
16 |
+
- length: int
|
17 |
+
- audio: str (オーディオファイルURL)
|
18 |
+
- title: str
|
19 |
+
|
20 |
+
タイトルリストファイルの例
|
21 |
+
|
22 |
+
<div>
|
23 |
+
<style scoped>
|
24 |
+
.dataframe tbody tr th:only-of-type {
|
25 |
+
vertical-align: middle;
|
26 |
+
}
|
27 |
+
|
28 |
+
.dataframe tbody tr th {
|
29 |
+
vertical-align: top;
|
30 |
+
}
|
31 |
+
|
32 |
+
.dataframe thead th {
|
33 |
+
text-align: right;
|
34 |
+
}
|
35 |
+
</style>
|
36 |
+
<table border="1" class="dataframe">
|
37 |
+
<thead>
|
38 |
+
<tr style="text-align: right;">
|
39 |
+
<th></th>
|
40 |
+
<th>id</th>
|
41 |
+
<th>date</th>
|
42 |
+
<th>length</th>
|
43 |
+
<th>audio</th>
|
44 |
+
<th>title</th>
|
45 |
+
</tr>
|
46 |
+
</thead>
|
47 |
+
<tbody>
|
48 |
+
<tr>
|
49 |
+
<th>0</th>
|
50 |
+
<td>69</td>
|
51 |
+
<td>2023-01-09</td>
|
52 |
+
<td>20993616</td>
|
53 |
+
<td>https://anchor.fm/s/14480e04/podcast/play/6323...</td>
|
54 |
+
<td>#69 2023年新年挨拶から 2022年の振り返りと2023年の抱負</td>
|
55 |
+
</tr>
|
56 |
+
<tr>
|
57 |
+
<th>1</th>
|
58 |
+
<td>70</td>
|
59 |
+
<td>2023-03-09</td>
|
60 |
+
<td>103287296</td>
|
61 |
+
<td>https://anchor.fm/s/14480e04/podcast/play/6621...</td>
|
62 |
+
<td>#70 PyCon JP Association代表理事退任と今後の展望をIqbalさんと語る</td>
|
63 |
+
</tr>
|
64 |
+
<tr>
|
65 |
+
<th>2</th>
|
66 |
+
<td>71</td>
|
67 |
+
<td>2023-03-22</td>
|
68 |
+
<td>116393694</td>
|
69 |
+
<td>https://anchor.fm/s/14480e04/podcast/play/6706...</td>
|
70 |
+
<td>#71 hirokikyさんをゲストに 自然言語処理系AI Chat GPT / Whisp...</td>
|
71 |
+
</tr>
|
72 |
+
<tr>
|
73 |
+
<th>3</th>
|
74 |
+
<td>72</td>
|
75 |
+
<td>2023-05-04</td>
|
76 |
+
<td>49642320</td>
|
77 |
+
<td>https://anchor.fm/s/14480e04/podcast/play/6976...</td>
|
78 |
+
<td>#72 PyCon US 2023 ひとり振り返り</td>
|
79 |
+
</tr>
|
80 |
+
<tr>
|
81 |
+
<th>4</th>
|
82 |
+
<td>73</td>
|
83 |
+
<td>2023-05-24</td>
|
84 |
+
<td>150643013</td>
|
85 |
+
<td>https://anchor.fm/s/14480e04/podcast/play/7094...</td>
|
86 |
+
<td>#73 Nyohoさんをゲストに Scratchからディープラーニングや数学の話</td>
|
87 |
+
</tr>
|
88 |
+
</tbody>
|
89 |
+
</table>
|
90 |
+
</div>
|
91 |
+
|
92 |
+
### 文字データ作成
|
93 |
+
|
94 |
+
- dataフォルダをを作る(srcと同じ階層)
|
95 |
+
- dataフォルダに、srtファイルを入れる
|
96 |
+
- (以下に従うと、srtファイルからIDが取得できる)
|
97 |
+
- 拡張子を `.srt` とする
|
98 |
+
- ファイル名に、ID(整数)が1つだけ入ってること
|
99 |
+
- IDの前後に、 `-` または `_` で区切られいること
|
100 |
+
- 以下のスクリプトを実行する。 `store` フォルダに `parquet` ファイルが srtファイル分できる
|
101 |
+
|
102 |
+
```
|
103 |
+
% python src/episode.py
|
104 |
+
```
|
105 |
+
|
106 |
+
### データベース作成
|
107 |
+
|
108 |
+
以下のコマンドで、テーブル作成から必要な3つのデータをDuckDB(永続化)を作る
|
109 |
+
|
110 |
+
```
|
111 |
+
% python src/store.py all
|
112 |
+
```
|
113 |
+
|
114 |
+
上記のコマンドの詳細
|
115 |
+
|
116 |
+
- テーブル作成 create table
|
117 |
+
- `python src/store.py create`
|
118 |
+
- タイトルリスト insert
|
119 |
+
- `python src/store.py podcastinsert`
|
120 |
+
- エピソードとテキスト insert
|
121 |
+
- `python src/store.py episodeinsert`
|
122 |
+
- ベクトル化 embedding
|
123 |
+
- `python src/store.py embed`
|
124 |
+
- ベクトルデータ index
|
125 |
+
- `python src/store.py index`
|
126 |
+
|
127 |
+
|
128 |
+
### 検索UI
|
129 |
+
|
130 |
+
```
|
131 |
+
% streamlit run src/app.py
|
132 |
+
```
|
133 |
+
|
134 |
+
- Podcastタイトル(複数)を選ぶ。未選択の場合すべてとなる
|
135 |
+
- 検索したいワードをテキストボックスに入力
|
136 |
+
- 10個のセンテンス(文章)候補が出てくる
|
137 |
+
- 表の左をクリックすると、下部に文字列が表示される
|
138 |
+
- 音声のタイミング(分・秒)が表示される・・未実装
|
139 |
+
- そのタイミングの音声がその場で聞ける・・将来的に実装したいが実現方法未確定
|
140 |
+
|
src/config.py
CHANGED
@@ -1,6 +1,14 @@
|
|
|
|
|
|
1 |
from pathlib import Path
|
2 |
# import logging
|
3 |
|
4 |
|
5 |
HERE = Path(__file__).resolve().parent
|
6 |
DUCKDB_FILE = HERE.parent / "db" / "terapyon-podcast.duckdb"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datetime import timedelta
|
2 |
+
import re
|
3 |
from pathlib import Path
|
4 |
# import logging
|
5 |
|
6 |
|
7 |
HERE = Path(__file__).resolve().parent
|
8 |
DUCKDB_FILE = HERE.parent / "db" / "terapyon-podcast.duckdb"
|
9 |
+
STORE_DIR = HERE.parent / "store"
|
10 |
+
DATA_DIR = HERE.parent / "data"
|
11 |
+
PODCAST_TITLE_LIST = str(STORE_DIR / 'title-list-202301-202501.parquet')
|
12 |
+
EPISODES_PARQUET = str(STORE_DIR / 'podcast-*.parquet')
|
13 |
+
divider_time = timedelta(minutes=1)
|
14 |
+
RE_PODCAST_SRT_FILE = re.compile(r"[_-](\d+)[_-]")
|
src/episode.py
CHANGED
@@ -1,23 +1,15 @@
|
|
1 |
from dataclasses import dataclass
|
2 |
from datetime import time as dt_time
|
3 |
from datetime import timedelta
|
4 |
-
from pathlib import Path
|
5 |
-
import re
|
6 |
import pandas as pd
|
7 |
-
|
8 |
-
|
9 |
-
HERE = Path(__file__).parent
|
10 |
-
DATA_DIR = HERE.parent / "data"
|
11 |
-
STORE_DIR = HERE.parent / "store"
|
12 |
-
divider_time = timedelta(minutes=5)
|
13 |
-
RE_PODCAST = re.compile(r"[_-](\d+)[_-]")
|
14 |
|
15 |
|
16 |
@dataclass
|
17 |
class SplitedText:
|
18 |
part: int
|
19 |
-
start:
|
20 |
-
end:
|
21 |
text: str
|
22 |
|
23 |
|
@@ -72,7 +64,10 @@ def make_episode(id_: int, title: str, srt_filename: str) -> Episode:
|
|
72 |
if start and second and text:
|
73 |
if abs(second - start) > divider_time:
|
74 |
end = second
|
75 |
-
st = SplitedText(part=part,
|
|
|
|
|
|
|
76 |
episode.texts.append(st)
|
77 |
|
78 |
# print(text)
|
@@ -95,7 +90,7 @@ def make_df(episode: Episode) -> pd.DataFrame:
|
|
95 |
def get_srt_files():
|
96 |
lst = []
|
97 |
for file_path in DATA_DIR.glob("*.srt"):
|
98 |
-
m =
|
99 |
if m is not None:
|
100 |
filename = file_path.name
|
101 |
id_ = int(m.group(1))
|
|
|
1 |
from dataclasses import dataclass
|
2 |
from datetime import time as dt_time
|
3 |
from datetime import timedelta
|
|
|
|
|
4 |
import pandas as pd
|
5 |
+
from config import STORE_DIR, DATA_DIR, divider_time, RE_PODCAST_SRT_FILE
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
|
8 |
@dataclass
|
9 |
class SplitedText:
|
10 |
part: int
|
11 |
+
start: int
|
12 |
+
end: int
|
13 |
text: str
|
14 |
|
15 |
|
|
|
64 |
if start and second and text:
|
65 |
if abs(second - start) > divider_time:
|
66 |
end = second
|
67 |
+
st = SplitedText(part=part,
|
68 |
+
start=int(start.total_seconds()),
|
69 |
+
end=int(end.total_seconds()),
|
70 |
+
text=text)
|
71 |
episode.texts.append(st)
|
72 |
|
73 |
# print(text)
|
|
|
90 |
def get_srt_files():
|
91 |
lst = []
|
92 |
for file_path in DATA_DIR.glob("*.srt"):
|
93 |
+
m = RE_PODCAST_SRT_FILE.search(file_path.name)
|
94 |
if m is not None:
|
95 |
filename = file_path.name
|
96 |
id_ = int(m.group(1))
|
src/store.py
CHANGED
@@ -1,17 +1,13 @@
|
|
1 |
-
from pathlib import Path
|
2 |
import duckdb
|
3 |
from embedding import get_embeddings
|
4 |
from config import DUCKDB_FILE
|
5 |
-
|
6 |
-
|
7 |
-
HERE = Path(__file__).parent
|
8 |
-
STORE_DIR = HERE.parent / "store"
|
9 |
|
10 |
|
11 |
def create_table():
|
12 |
conn = duckdb.connect(DUCKDB_FILE)
|
13 |
podcasts_create = """CREATE TABLE podcasts (
|
14 |
-
id BIGINT PRIMARY KEY,
|
15 |
title TEXT, date DATE, guests TEXT[], length BIGINT, audio TEXT
|
16 |
);
|
17 |
"""
|
@@ -39,7 +35,7 @@ def insert_podcast():
|
|
39 |
SELECT id, title, date, [], length, audio
|
40 |
FROM read_parquet(?);
|
41 |
"""
|
42 |
-
conn.execute(sql, [
|
43 |
conn.commit()
|
44 |
conn.close()
|
45 |
|
@@ -50,7 +46,7 @@ def insert_episodes():
|
|
50 |
SELECT id, part, start, end_, text
|
51 |
FROM read_parquet(?);
|
52 |
"""
|
53 |
-
conn.execute(sql, [
|
54 |
conn.commit()
|
55 |
conn.close()
|
56 |
|
|
|
|
|
1 |
import duckdb
|
2 |
from embedding import get_embeddings
|
3 |
from config import DUCKDB_FILE
|
4 |
+
from config import PODCAST_TITLE_LIST, EPISODES_PARQUET
|
|
|
|
|
|
|
5 |
|
6 |
|
7 |
def create_table():
|
8 |
conn = duckdb.connect(DUCKDB_FILE)
|
9 |
podcasts_create = """CREATE TABLE podcasts (
|
10 |
+
id BIGINT PRIMARY KEY,
|
11 |
title TEXT, date DATE, guests TEXT[], length BIGINT, audio TEXT
|
12 |
);
|
13 |
"""
|
|
|
35 |
SELECT id, title, date, [], length, audio
|
36 |
FROM read_parquet(?);
|
37 |
"""
|
38 |
+
conn.execute(sql, [PODCAST_TITLE_LIST])
|
39 |
conn.commit()
|
40 |
conn.close()
|
41 |
|
|
|
46 |
SELECT id, part, start, end_, text
|
47 |
FROM read_parquet(?);
|
48 |
"""
|
49 |
+
conn.execute(sql, [EPISODES_PARQUET])
|
50 |
conn.commit()
|
51 |
conn.close()
|
52 |
|