SynthBanshee/pyproject.toml at main · DataHackIL/SynthBanshee · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "synthbanshee"
version = "0.1.0"
description = "SynthBanshee: synthetic Hebrew audio dataset generator for the AVDP initiative"
readme = "README.md"
requires-python = ">=3.11"
license = { text = "MIT" }

dependencies = [
    # --- Config & validation ---
    "pydantic>=2.6",           # Config models, label schema validation
    "pyyaml>=6.0",             # YAML config loading
    "jinja2>=3.1",             # Scene script templates

    # --- Audio processing ---
    "soundfile>=0.12",         # WAV I/O (libsndfile bindings)
    "pydub>=0.25",             # Audio manipulation, mixing, segment slicing
    "pyroomacoustics>=0.7",    # Room impulse response simulation (Tier B)
    "numpy>=1.26",
    "scipy>=1.12",             # Butterworth filter, spectral subtraction

    # --- TTS providers ---
    "azure-cognitiveservices-speech>=1.36",   # Azure TTS he-IL (primary)

    # --- LLM script generation ---
    "anthropic>=0.25",         # Claude API for script generation
    "openai>=1.30",            # GPT-4 fallback for script generation

    # --- Metadata & I/O ---
    "pandas>=2.2",             # Manifest CSV generation and QA
    "jsonlines>=4.0",          # JSONL label file I/O

    # --- Hashing & caching ---
    "joblib>=1.3",             # TTS render cache (memory-mapped)
    "xxhash>=3.4",             # Fast content hashing for cache keys

    # --- CLI ---
    "click>=8.1",              # CLI entry points
    "rich>=13.7",              # Progress bars, console output

    # --- Dataset QA ---
    "librosa>=0.10",           # Audio analysis, SNR estimation, spectrogram QA
    "pyloudnorm>=0.1",         # ITU-R BS.1770-4 loudness measurement (M10a)
]

[project.optional-dependencies]
# Tier C (confusor) soundscape synthesis — requires native sox headers (brew install sox)
tier-c = [
    "scaper>=1.6",
]

# Google Cloud TTS — evaluation/secondary provider only
google-tts = [
    "google-cloud-texttospeech>=2.16",
]

# Whisper-based ASR sanity for `qa-report --asr` (#87 length-ratio detector).
# Heavyweight: pulls torch + transformers + ~3 GB of Whisper-large-v3 weights
# on first use.  Not required for normal generation/QA — only install when
# running the ASR sanity check locally (see CLAUDE.md "ASR sanity check policy").
eval-asr = [
    "torch>=2.0",
    "transformers>=4.40",
    "accelerate>=0.30",
    "jiwer>=3.0",
]

dev = [
    "pytest>=8.0",
    "pytest-cov>=5.0",
    "pytest-asyncio>=0.23",
    "ruff>=0.4",               # Linting + formatting
    "mypy>=1.9",
    "ipykernel>=6.29",         # Notebook QA
    "matplotlib>=3.8",         # QA visualisations
    "seaborn>=0.13",
]

[project.scripts]
synthbanshee = "synthbanshee.cli:cli"

[tool.hatch.build.targets.wheel]
packages = ["synthbanshee"]

[tool.ruff]
line-length = 100
target-version = "py311"

[tool.ruff.lint]
select = ["E", "F", "I", "UP", "B", "SIM"]
ignore = ["E501"]

[tool.mypy]
python_version = "3.11"
strict = false
ignore_missing_imports = true

[tool.pytest.ini_options]
testpaths = ["tests"]
asyncio_mode = "auto"
markers = [
    "live_tts: integration tests that hit a real TTS API (skip with -m 'not live_tts')",
]

[tool.coverage.run]
source = ["synthbanshee"]
relative_files = true

[tool.coverage.paths]
source = [
    "synthbanshee",
    "/home/runner/work/*/*/synthbanshee",
]

[tool.coverage.report]
show_missing = true

[tool.coverage.xml]
output = "coverage.xml"