vcref 0.1.0__tar.gz → 0.2.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {vcref-0.1.0 → vcref-0.2.0}/PKG-INFO +12 -1
- {vcref-0.1.0 → vcref-0.2.0}/README.md +11 -0
- {vcref-0.1.0 → vcref-0.2.0}/pyproject.toml +1 -1
- {vcref-0.1.0 → vcref-0.2.0}/src/vcref/__init__.py +1 -1
- {vcref-0.1.0 → vcref-0.2.0}/src/vcref/bundle.py +40 -0
- {vcref-0.1.0 → vcref-0.2.0}/tests/test_bundle.py +20 -0
- {vcref-0.1.0 → vcref-0.2.0}/.gitignore +0 -0
- {vcref-0.1.0 → vcref-0.2.0}/LICENSE +0 -0
- {vcref-0.1.0 → vcref-0.2.0}/src/vcref/__main__.py +0 -0
- {vcref-0.1.0 → vcref-0.2.0}/src/vcref/cli.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: vcref
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.2.0
|
|
4
4
|
Summary: Read, write and verify .vcref voice reference bundles for zero-shot voice-cloning TTS.
|
|
5
5
|
Project-URL: Homepage, https://github.com/bhavya091213/vcref
|
|
6
6
|
License: MIT
|
|
@@ -44,3 +44,14 @@ vcref.set_cadence("voice.vcref", cadence_dict) # rewrite only the cadence block
|
|
|
44
44
|
```
|
|
45
45
|
|
|
46
46
|
Errors are `vcref.VcrefError`. Command line: `vcref info|verify <file>`, `vcref unpack <file> <dir> [--wav]`.
|
|
47
|
+
|
|
48
|
+
## Choosing and inspecting references
|
|
49
|
+
|
|
50
|
+
```python
|
|
51
|
+
b.explain("Really?") # {'tag', 'why': 'rule'|'default', 'rule', 'pattern'}
|
|
52
|
+
b.prompt("Really?", stack=True) # [selected clip, default clip] ready for a TTS prompt
|
|
53
|
+
b.search("podcast") # tags whose tag, label or transcript contain the text
|
|
54
|
+
b.stats() # clip/part bytes, kbps, pcm16_ratio, per-clip sizes (no decoding)
|
|
55
|
+
```
|
|
56
|
+
|
|
57
|
+
Guides: [compression](../docs/compression.md), [extracting references](../docs/extracting-references.md), [best practices](../docs/best-practices.md).
|
|
@@ -27,3 +27,14 @@ vcref.set_cadence("voice.vcref", cadence_dict) # rewrite only the cadence block
|
|
|
27
27
|
```
|
|
28
28
|
|
|
29
29
|
Errors are `vcref.VcrefError`. Command line: `vcref info|verify <file>`, `vcref unpack <file> <dir> [--wav]`.
|
|
30
|
+
|
|
31
|
+
## Choosing and inspecting references
|
|
32
|
+
|
|
33
|
+
```python
|
|
34
|
+
b.explain("Really?") # {'tag', 'why': 'rule'|'default', 'rule', 'pattern'}
|
|
35
|
+
b.prompt("Really?", stack=True) # [selected clip, default clip] ready for a TTS prompt
|
|
36
|
+
b.search("podcast") # tags whose tag, label or transcript contain the text
|
|
37
|
+
b.stats() # clip/part bytes, kbps, pcm16_ratio, per-clip sizes (no decoding)
|
|
38
|
+
```
|
|
39
|
+
|
|
40
|
+
Guides: [compression](../docs/compression.md), [extracting references](../docs/extracting-references.md), [best practices](../docs/best-practices.md).
|
|
@@ -169,6 +169,46 @@ class Bundle:
|
|
|
169
169
|
return tag
|
|
170
170
|
return self.default
|
|
171
171
|
|
|
172
|
+
def explain(self, text):
|
|
173
|
+
"""Why `select_tag(text)` returns what it does: {"tag", "why": "rule"|"default", "rule", "pattern"}."""
|
|
174
|
+
for i, (rx, tag) in enumerate(self.rules()):
|
|
175
|
+
if rx.search(text) and tag in self.manifest["clips"]:
|
|
176
|
+
return {"tag": tag, "why": "rule", "rule": i, "pattern": rx.pattern}
|
|
177
|
+
return {"tag": self.default, "why": "default", "rule": None, "pattern": None}
|
|
178
|
+
|
|
179
|
+
def prompt(self, text, stack=False):
|
|
180
|
+
"""Reference clips to hand a TTS model for `text`: the rule-selected clip first, then (stack=True) the
|
|
181
|
+
default clip as a second, neutral anchor when it differs."""
|
|
182
|
+
tag = self.select_tag(text)
|
|
183
|
+
clips = [self.clip(tag)]
|
|
184
|
+
if stack and tag != self.default:
|
|
185
|
+
clips.append(self.clip(self.default))
|
|
186
|
+
return clips
|
|
187
|
+
|
|
188
|
+
def search(self, query):
|
|
189
|
+
"""Tags whose tag, label or transcript contains `query` (case-insensitive), in manifest order."""
|
|
190
|
+
q = query.lower()
|
|
191
|
+
return [t for t, c in self.manifest["clips"].items()
|
|
192
|
+
if q in t.lower() or q in (c.get("label") or "").lower() or q in c["text"].lower()]
|
|
193
|
+
|
|
194
|
+
def stats(self):
|
|
195
|
+
"""Sizes and bitrates from the zip directory (no audio decoding). bytes are stored sizes; `pcm16_ratio`
|
|
196
|
+
compares the clips with the same audio as raw 16-bit PCM at each clip's own sample rate."""
|
|
197
|
+
per, raw = [], 0
|
|
198
|
+
for tag, c in self.manifest["clips"].items():
|
|
199
|
+
size = self._z.getinfo(c["file"]).file_size
|
|
200
|
+
secs, sr = c.get("seconds"), c.get("sample_rate")
|
|
201
|
+
if secs and sr:
|
|
202
|
+
raw += int(secs * sr * 2)
|
|
203
|
+
per.append({"tag": tag, "seconds": secs, "sample_rate": sr, "bytes": size,
|
|
204
|
+
"kbps": round(size * 8 / secs / 1000) if secs else None})
|
|
205
|
+
clip_bytes = sum(p["bytes"] for p in per)
|
|
206
|
+
part_bytes = sum(self._z.getinfo(p["file"]).file_size for p in self.manifest.get("parts", {}).values())
|
|
207
|
+
secs = sum(p["seconds"] or 0 for p in per)
|
|
208
|
+
return {"clips": len(per), "clip_seconds": round(secs, 1), "clip_bytes": clip_bytes, "part_bytes": part_bytes,
|
|
209
|
+
"file_bytes": clip_bytes + part_bytes, "kbps": round(clip_bytes * 8 / secs / 1000) if secs else None,
|
|
210
|
+
"pcm16_ratio": round(raw / clip_bytes, 2) if raw and clip_bytes else None, "per_clip": per}
|
|
211
|
+
|
|
172
212
|
def clip(self, tag=None):
|
|
173
213
|
"""Clip for `tag` (unknown or None -> default) with its FLAC bytes."""
|
|
174
214
|
if tag not in self.manifest["clips"]:
|
|
@@ -36,6 +36,26 @@ class FixtureTests(unittest.TestCase):
|
|
|
36
36
|
self.assertAlmostEqual(len(x) / sr, 0.5, places=2)
|
|
37
37
|
|
|
38
38
|
|
|
39
|
+
class HelperTests(unittest.TestCase):
|
|
40
|
+
def test_explain_and_prompt(self):
|
|
41
|
+
with vcref.Bundle(FIX) as b:
|
|
42
|
+
e = b.explain("Really?")
|
|
43
|
+
self.assertEqual((e["tag"], e["why"], e["rule"]), ("question", "rule", 0))
|
|
44
|
+
self.assertEqual(b.explain("Plain.")["why"], "default")
|
|
45
|
+
self.assertEqual([c.tag for c in b.prompt("Really?")], ["question"])
|
|
46
|
+
self.assertEqual([c.tag for c in b.prompt("Really?", stack=True)], ["question", "neutral"])
|
|
47
|
+
self.assertEqual([c.tag for c in b.prompt("Plain.", stack=True)], ["neutral"])
|
|
48
|
+
|
|
49
|
+
def test_search_and_stats(self):
|
|
50
|
+
with vcref.Bundle(FIX) as b:
|
|
51
|
+
self.assertIn("question", b.search("QUEST"))
|
|
52
|
+
s = b.stats()
|
|
53
|
+
self.assertEqual(s["clips"], 2)
|
|
54
|
+
self.assertEqual(s["file_bytes"], s["clip_bytes"] + s["part_bytes"])
|
|
55
|
+
self.assertEqual(len(s["per_clip"]), 2)
|
|
56
|
+
self.assertGreater(s["clip_bytes"], 0)
|
|
57
|
+
|
|
58
|
+
|
|
39
59
|
class RoundTripTests(unittest.TestCase):
|
|
40
60
|
def setUp(self):
|
|
41
61
|
self.tmp = Path(tempfile.mkdtemp())
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|