Spaces:
Sleeping
Sleeping
File size: 17,934 Bytes
a779b8d 8e0d6f9 9062c1f 8e0d6f9 37c155a 8e0d6f9 602513b 8e0d6f9 c79c7f8 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 b8c09e4 8e0d6f9 1c1f57b 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 9062c1f 8e0d6f9 7972254 8e0d6f9 9062c1f 8e0d6f9 c322ac9 2528c00 8e0d6f9 4f63d84 8e0d6f9 4f63d84 8e0d6f9 4f63d84 5cb2685 4f63d84 8e0d6f9 4f63d84 8e0d6f9 4f63d84 8e0d6f9 e92ae9b 8e0d6f9 e92ae9b 8e0d6f9 e92ae9b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 |
import streamlit as st
st.set_page_config(page_title="CantusSVS", layout="wide")
import os
import yaml
import shutil
import traceback
import json
import requests
import zipfile
import streamlit.components.v1 as components
from pathlib import Path
from webapp.services.defaults.default_splitter import split_syllable
def patch_config_yaml_files():
root = "/tmp/cantussvs_v1"
checkpoints_root = os.path.join(root, "checkpoints")
data_root = os.path.join(root, "data")
for dirpath, _, filenames in os.walk(checkpoints_root):
for filename in filenames:
if filename == "config.yaml":
full_path = os.path.join(dirpath, filename)
try:
with open(full_path, "r") as f:
config = yaml.safe_load(f)
if not isinstance(config, dict):
continue
modified = False
for key, value in config.items():
if isinstance(value, str):
if value.startswith("checkpoints/"):
rel = value.split("/", 1)[1]
config[key] = os.path.join(checkpoints_root, rel)
modified = True
elif value.startswith("data/"):
rel = value.split("/", 1)[1]
config[key] = os.path.join(data_root, rel)
modified = True
if modified:
with open(full_path, "w") as f:
yaml.dump(config, f)
print(f"✅ Patched paths in {full_path}")
except Exception as e:
print(f"❌ Failed to patch {full_path}: {e}")
# Disable Streamlit file watcher
os.environ['STREAMLIT_SERVER_FILE_WATCHER_TYPE'] = 'none'
# Ensure project root is on the import path
PROJECT_ROOT = Path(__file__).resolve().parent
import sys
sys.path.insert(0, str(PROJECT_ROOT))
from webapp.services.parsing.mei_parser import parse_mei_for_editor
from webapp.services.parsing.ds_builder import build_ds_from_notes
from webapp.services.parsing.ds_validator import validate_ds
from webapp.services.phonemes.phoneme_dict import PHONEMES as permitted_phonemes
from inference.pipeline import run_inference
def safe_symlink(src, dst):
try:
if os.path.islink(dst):
if os.readlink(dst) == src:
print(f"✅ Symlink already correct: {dst} → {src}")
return
else:
print(f"⚠️ Symlink exists but points elsewhere. Skipping: {dst}")
return
elif os.path.exists(dst):
print(f"❗ Cannot create symlink, path exists and is not a symlink: {dst}")
return
os.symlink(src, dst)
print(f"✅ Created symlink: {dst} → {src}")
except Exception as e:
print(f"❗ Failed to create symlink {dst} -> {src}: {e}")
# Directories
HF_CHECKPOINTS_DIR = "/tmp/cantussvs_v1/checkpoints"
HF_DATA_DIR = "/tmp/cantussvs_v1/data"
DEMO_FILES = PROJECT_ROOT / "webapp/demo_files"
UPLOAD_MEI_DIR = PROJECT_ROOT / "webapp/uploaded_mei"
UPLOAD_DS_DIR = PROJECT_ROOT / "webapp/uploaded_ds"
TMP_DS_DIR = PROJECT_ROOT / "webapp/tmp_ds"
OUTPUT_DIR = PROJECT_ROOT / "webapp/output"
for d in [DEMO_FILES, UPLOAD_MEI_DIR, UPLOAD_DS_DIR, TMP_DS_DIR, OUTPUT_DIR]:
d.mkdir(parents=True, exist_ok=True)
@st.cache_resource
def download_and_extract_from_hf():
url = "https://huggingface.co/datasets/liampond/CantusSVS/resolve/main/cantussvs_v1.zip"
zip_path = "/tmp/cantussvs_v1.zip"
extract_dir = "/tmp/cantussvs_v1"
if not os.path.exists(extract_dir):
st.write("📦 Downloading data + model from Hugging Face...")
r = requests.get(url, stream=True)
with open(zip_path, "wb") as f:
for chunk in r.iter_content(chunk_size=8192):
f.write(chunk)
st.write("📂 Extracting contents...")
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
zip_ref.extractall(extract_dir)
# ✅ Only do this once, right after unzip
patch_config_yaml_files()
safe_symlink(os.path.join(extract_dir, "checkpoints"), "checkpoints")
safe_symlink(os.path.join(extract_dir, "data"), "data")
return extract_dir
# Call it once and use it globally
base_path = download_and_extract_from_hf()
patch_config_yaml_files()
st.write("✅ Loaded assets to:", base_path)
# CSS styling
# st.markdown("""
# <style>
# html, body, [class*="css"] { font-size: 18px !important; }
# div[data-testid="stSelectbox"] label,
# div[data-testid="stNumberInput"] label,
# div[data-testid="stTextInput"] label { font-size: 13px; padding-bottom: 0px; }
# div[data-testid="stSlider"] label { font-size: 0px; }
# div.stButton > button:first-child {
# background-color: black; color: white;
# font-size: 14px; padding: 4px 10px;
# border-radius: 8px;
# }
# section[data-testid="stFileUploaderDropzone"] { padding: 2rem; }
# .tooltip {
# position: relative;
# display: inline-block;
# border-bottom: 1px dotted white;
# cursor: help;
# }
# .tooltip .tooltiptext {
# visibility: hidden;
# width: 250px;
# background-color: black;
# color: #fff;
# text-align: center;
# border-radius: 6px;
# padding: 6px;
# position: absolute;
# z-index: 1;
# bottom: 125%; /* Position above */
# left: 50%;
# margin-left: -125px;
# opacity: 0;
# transition: opacity 0.3s;
# }
# .tooltip:hover .tooltiptext {
# visibility: visible;
# opacity: 1;
# }
# </style>
# """, unsafe_allow_html=True)
# Phoneme mappings
phoneme_display_map = { "ap": "Pause", "br": "Breath" }
display_to_phoneme = {v: k for k, v in phoneme_display_map.items()}
full_phoneme_list_display = [phoneme_display_map.get(p, p) for p in permitted_phonemes]
# Pitch list D4-D5
allowed_pitches = ["D4", "D#4", "E4", "F4", "F#4", "G4", "G#4", "A4", "A#4", "B4", "C5", "C#5", "D5"]
# Title
st.title("CantusSVS: Latin Singing Voice Synthesis")
st.markdown("""
# About CantusSVS
<p>CantusSVS is a web-based Singing Voice Synthesis (SVS) system designed for composers and musicians to synthesize Latin chant audio from a custom musical score.
Built on top of the DiffSinger AI model, CantusSVS enables detailed, precise control over melody, rhythm, phonemes, and timing without any programming knowledge required.</p>
<p>Designed by Liam Pond as the final project for MUS6329X: Projet en informatique musicale (Prof. Dominic Thibault) at the Université de Montréal. For more information, you can view the README.md under the 'Files' tab of this Space.</p>
You can find DiffSinger in the following paper:
**DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism**
Liu, Jinglin, Chengxi Li, Yi Ren, Feiyang Chen, and Zhou Zhao. 2022. "Diffsinger: Singing Voice Synthesis via Shallow Diffusion Mechanism." In *Proceedings of the AAAI Conference on Artificial Intelligence* 36 10: 11020–11028. [https://arxiv.org/abs/2105.02446](http://dx.doi.org/10.1609/aaai.v36i10.21350).
Model training was done using Cedar, a cluster provided by the Digital Research Alliance of Canada. To train your own model locally, follow [this tutorial](https://youtu.be/Sxt11TAflV0?feature=shared) by [tigermeat](https://www.youtube.com/@spicytigermeat).
For general help training and creating a dataset, [this tutorial](https://docs.google.com/document/d/1uMsepxbdUW65PfIWL1pt2OM6ZKa5ybTTJOpZ733Ht6s/view) by [PixPrucer](https://bsky.app/profile/pixprucer.bsky.social) is an excellent guide. For help, join the [DiffSinger Discord server](https://discord.gg/DZ6fhEUfnb).
The dataset used for this project was built using [*Adventus: Dominica prima adventus Domini*](https://youtu.be/ThnPySybDJs?feature=shared), the first track from [Psallentes](https://psallentes.com/)' album *Salzinnes Saints*. Psallentes is a Belgian women's chorus that specializes in Late Medieval and Renaissance music. *Salzinnes Saints* is an album of music from the [Salzinnes Antiphonal](https://www.smu.ca/academics/archives/the-salzinnes-antiphonal.html), a mid-sixteenth century choirbook with the music and text for the Liturgy of the Hours.
---
# How to Use CantusSVS
## 1. Compose Your Music
Compose the chant you want to synthesize using the notation software of your choice. [MuseScore 4](https://musescore.org/en/download) is recommended.
The chant must adhere to the following conditions:
- Monophonic only (one note at a time, no harmonies or chords)
- Pitch range of <span class="tooltip">**D4 to D5**<span class="tooltiptext">Because training data was limited outside this range, synthesis outside these pitches is very poor.</span></span>
- Lyrics (Latin) under each note, separated by syllable
## 2. Export Your Score to MEI
When your score is complete, export it to MEI.
In MuseScore:
- Go to **File → Export**
- Choose the `.mei` file format
- Save it to your computer
## 3. Upload Your Score to CantusSVS
In the CantusSVS web app:
- Select **MEI** mode
- Adjust the **tempo** if necessary using the provided slider
- Upload your `.mei` file
- Your score will be displayed using Verovio
- You may use the demo `.mei` file if you wish
## 4. Edit Phonemes, Durations, and Pitches
CantusSVS automatically suggests phoneme splits for each syllable.
However, you will have the opportunity to review phonemes, durations, and pitches.
## 5. Synthesize the Audio
When you're done:
- Click **Confirm**
- CantusSVS will create a `.ds` file which are processed through pretrained DiffSinger models
- The synthesized chant will be generated
This can take a few minutes depending on input length
## 6. Listen and Download
After synthesis you can either listen to your chant directly in the app or download a `.wav` file to your computer.
---
""", unsafe_allow_html=True)
st.markdown("""
<script>
const tooltipSpan = window.parent.document.querySelector('span[style*="border-bottom: 1px dotted black"]');
if (tooltipSpan) {
tooltipSpan.addEventListener('mouseover', () => {
tooltipSpan.children[0].style.visibility = 'visible';
tooltipSpan.children[0].style.opacity = 1;
});
tooltipSpan.addEventListener('mouseout', () => {
tooltipSpan.children[0].style.visibility = 'hidden';
tooltipSpan.children[0].style.opacity = 0;
});
}
</script>
""", unsafe_allow_html=True)
filetype = st.selectbox("Select file type:", ["MEI", "DS"])
def handle_exception(context_message):
st.error(f"{context_message}. See console.")
print("\n" + "="*30)
print(f"Exception during {context_message}")
traceback.print_exc()
print("="*30 + "\n")
st.stop()
if filetype == "MEI":
st.header("1. Select MEI Source")
use_demo = st.checkbox("Use demo MEI file", value=False)
tempo = st.slider("Tempo (BPM)", 1, 300, 60)
if use_demo:
mei_path = DEMO_FILES / "Demo1.mei"
if not mei_path.exists():
st.error("Demo MEI file missing.")
st.stop()
with open(mei_path, "rb") as f:
mei_file_bytes = f.read()
else:
mei_file = st.file_uploader("Upload your MEI file", type="mei")
if not mei_file:
st.stop()
mei_path = UPLOAD_MEI_DIR / mei_file.name
with open(mei_path, "wb") as f:
f.write(mei_file.getbuffer())
mei_file_bytes = mei_file.getvalue()
mei_text = mei_file_bytes.decode("utf-8")
try:
raw_notes = parse_mei_for_editor(mei_path, tempo)
except Exception:
handle_exception("MEI parsing")
# Always update session state
st.session_state.original_raw_notes = raw_notes
syllable_groups = []
for note in st.session_state.original_raw_notes:
syllable_text = note["lyric"]
pitch = note["pitch"]
syllable = split_syllable(
syllable=syllable_text,
note_duration=note["duration"],
tempo=tempo,
pitch=pitch
)
syllable_groups.append({
"syllable": syllable_text,
"phonemes": syllable
})
if "edited_syllables" not in st.session_state:
st.session_state.edited_syllables = syllable_groups
st.subheader("Score Preview")
components.html(f"""<div id=\"app\" style=\"border: 1px solid lightgray; min-height: 400px;\"></div><script type=\"module\">import 'https://editor.verovio.org/javascript/app/verovio-app.js';const app=new Verovio.App(document.getElementById(\"app\"),{{defaultView:'document',documentZoom:4}});app.loadData(`{mei_text}`);</script>""", height=500)
st.header("2. Edit Phonemes, Durations, and Pitches")
updated_syllables = []
if "previous_tempo" not in st.session_state:
st.session_state.previous_tempo = tempo
if tempo != st.session_state.previous_tempo:
for i, note in enumerate(st.session_state.original_raw_notes):
updated = split_syllable(
syllable=note["lyric"],
note_duration=note["duration"],
tempo=tempo,
pitch=note["pitch"]
)
# preserve existing phoneme values (if possible)
for j, ph in enumerate(updated):
try:
existing = st.session_state.edited_syllables[i]["phonemes"][j]
ph["phoneme"] = existing["phoneme"]
ph["pitch"] = existing["pitch"]
except IndexError:
pass # new phoneme or longer split
st.session_state.edited_syllables[i]["phonemes"] = updated
st.session_state.previous_tempo = tempo
for idx, group in enumerate(st.session_state.edited_syllables):
st.markdown(f"#### {group['syllable'].capitalize()}")
new_phonemes = []
for j, ph in enumerate(group["phonemes"]):
col1, col2, col3, col4 = st.columns([3, 3, 3, 1]) # new column for delete button
with col1:
phoneme_display = st.selectbox(
"Phoneme",
full_phoneme_list_display,
index=full_phoneme_list_display.index(phoneme_display_map.get(ph["phoneme"], ph["phoneme"])),
key=f"phoneme_{idx}_{j}"
)
phoneme_internal = display_to_phoneme.get(phoneme_display, phoneme_display)
with col2:
duration = st.number_input(
"Duration (seconds)",
min_value=0.0, max_value=5.0,
value=float(ph["duration"]),
step=0.01, format="%.2f",
key=f"duration_num_{idx}_{j}"
)
with col3:
pitch = st.selectbox(
"Pitch",
allowed_pitches,
index=allowed_pitches.index(ph["pitch"]) if ph["pitch"] in allowed_pitches else 0,
key=f"pitch_{idx}_{j}"
)
with col4:
if st.button("❌", key=f"remove_{idx}_{j}"):
group["phonemes"].pop(j)
st.experimental_rerun() # force rerender safely
new_phonemes.append({"phoneme": phoneme_internal, "duration": duration, "pitch": pitch})
if st.button("➕ Add Phoneme", key=f"add_phoneme_{idx}"):
group["phonemes"].append({"phoneme": "a", "duration": 0.2, "pitch": "D4"})
st.experimental_rerun()
updated_syllables.append({"syllable": group["syllable"], "phonemes": new_phonemes})
st.divider()
st.session_state.edited_syllables = updated_syllables
st.header("3. Synthesize")
confirm_clicked = st.button("✅ Synthesize", key="confirm_button_mei")
if confirm_clicked:
ds_path = TMP_DS_DIR / f"{mei_path.stem}.ds"
try:
all_phonemes = [ph for syllable in st.session_state.edited_syllables for ph in syllable["phonemes"]]
build_ds_from_notes(all_phonemes, ds_path)
with open(ds_path, "r", encoding="utf-8") as f:
ds_data = json.load(f)
validate_ds(ds_data)
st.success(f"DS file created: {ds_path.name}")
except Exception:
handle_exception("DS generation or validation")
with st.spinner("Running DiffSinger inference…"):
try:
wav_path = run_inference(ds_path, OUTPUT_DIR, mei_path.stem)
except Exception:
handle_exception("inference")
st.success("Synthesis complete!")
st.audio(str(wav_path))
st.download_button("Download WAV", data=open(wav_path, "rb"), file_name=wav_path.name)
elif filetype == "DS":
st.header("1. Upload DS File")
ds_file = st.file_uploader("Upload your .ds file", type=["ds", "json"])
st.header("2. Synthesize")
synth_clicked = st.button("✅ Synthesize", key="synthesize_button_ds")
if synth_clicked:
if not ds_file:
st.error("Please upload a .ds file.")
st.stop()
ds_path = UPLOAD_DS_DIR / ds_file.name
with open(ds_path, "wb") as f:
f.write(ds_file.getbuffer())
with open(ds_path, "r", encoding="utf-8") as f:
ds_data = json.load(f)
try:
validate_ds(ds_data)
except Exception as e:
st.error(f"Invalid DS file: {e}")
st.stop()
with st.spinner("Running DiffSinger inference…"):
try:
wav_path = run_inference(ds_path, OUTPUT_DIR, ds_path.stem)
except Exception:
handle_exception("inference")
st.success("Synthesis complete!")
st.audio(str(wav_path))
st.download_button("Download WAV", data=open(wav_path, "rb"), file_name=wav_path.name) |