Merge pull request #15 from descriptinc/ps/discourse

Discourse upload, widget improvements, bug fixes.

.github/workflows/pr.yml

@@ -8,7 +8,7 @@ jobs:
       - uses: actions/checkout@v2
       # Run isort + black formatter
       - name: Python Code Formatter
-        uses: tdeboissiere/python-format-action@master
+        uses: descriptinc/python-format-action@master
 
   build:
     runs-on: ubuntu-latest

.gitignore

@@ -130,3 +130,4 @@ dmypy.json
 profile_speed.py.lprof
 
 scratch/
+*.ipynb

.pre-commit-config.yaml

@@ -4,12 +4,12 @@ repos:
   hooks:
     - id: reorder-python-imports
 - repo: https://github.com/psf/black
-  rev: 21.4b2
+  rev: 22.1.0
   hooks:
     - id: black
       language_version: python3
 - repo: https://github.com/pre-commit/pre-commit-hooks
-  rev: v3.4.0
+  rev: v4.0.1
   hooks:
     - id: end-of-file-fixer
     - id: trailing-whitespace

audiotools/__init__.py

@@ -1,3 +1,3 @@
-__version__ = "0.1.4"
+__version__ = "0.1.5"
 from .core import AudioSignal, STFTParams, Meter, util
 from . import metrics

audiotools/core/display.py

@@ -1,18 +1,30 @@
+import json
+import os
+import shlex
+import subprocess
+import tempfile
+
+import matplotlib.pyplot as plt
 import numpy as np
+from matplotlib.gridspec import GridSpec
 
 
 class DisplayMixin:
     def specshow(self, batch_idx=0, x_axis="time", y_axis="linear", **kwargs):
         import librosa
         import librosa.display
 
+        # Always re-compute the STFT data before showing it, in case
+        # it changed.
+        self.stft_data = None
+
         log_mag = librosa.amplitude_to_db(self.magnitude.cpu().numpy(), ref=np.max)
         librosa.display.specshow(
             log_mag[batch_idx].mean(axis=0),
             x_axis=x_axis,
             y_axis=y_axis,
             sr=self.sample_rate,
-            **kwargs
+            **kwargs,
         )
 
     def waveplot(self, batch_idx=0, x_axis="time", **kwargs):
@@ -24,3 +36,51 @@ def waveplot(self, batch_idx=0, x_axis="time", **kwargs):
         librosa.display.waveplot(
             audio_data, x_axis=x_axis, sr=self.sample_rate, **kwargs
         )
+
+    def wavespec(self, batch_idx=0, x_axis="time", **kwargs):
+        gs = GridSpec(6, 1)
+        plt.subplot(gs[0, :])
+        self.waveplot(batch_idx=batch_idx, x_axis=x_axis)
+        plt.subplot(gs[1:, :])
+        self.specshow(batch_idx=batch_idx, x_axis=x_axis, **kwargs)
+
+    def upload_to_discourse(
+        self,
+        label=None,
+        api_username=None,
+        api_key=None,
+        batch_idx=0,
+        discourse_server=None,
+        ext=".wav",
+    ):  # pragma: no cover
+        if api_username is None:
+            api_username = os.environ.get("DISCOURSE_API_USERNAME", None)
+        if api_key is None:
+            api_key = os.environ.get("DISCOURSE_API_KEY", None)
+        if discourse_server is None:
+            discourse_server = os.environ.get("DISCOURSE_SERVER", None)
+
+        if discourse_server is None or api_key is None or api_username is None:
+            raise RuntimeError(
+                "DISCOURSE_API_KEY, DISCOURSE_SERVER, DISCOURSE_API_USERNAME must be set in your environment!"
+            )
+
+        with tempfile.NamedTemporaryFile(suffix=ext) as f:
+            self.write(f.name, batch_idx=batch_idx)
+
+            command = (
+                f"curl -s -X POST {discourse_server}/uploads.json "
+                f"-H 'content-type: multipart/form-data;' "
+                f"-H 'Api-Key: {api_key}' "
+                f"-H 'Api-Username: {api_username}' "
+                f"-F 'type=composer' "
+                f"-F 'files[]=@{f.name}' "
+            )
+            info = json.loads(subprocess.check_output(shlex.split(command)))
+
+            label = self.path_to_input_file if label is None else label
+            if label is None:
+                label = "unknown"
+
+            formatted = f"![{label}|audio]({info['short_path']})"
+            return formatted, info

audiotools/core/dsp.py

@@ -121,7 +121,10 @@ def low_pass(self, cutoffs, zeros=51):
         cutoffs = util.ensure_tensor(cutoffs, 2, self.batch_size)
         cutoffs = cutoffs / self.sample_rate
         filtered = torch.empty_like(self.audio_data)
+
         for i, cutoff in enumerate(cutoffs):
-            filtered[i] = julius.lowpass_filter(self.audio_data[i], cutoff, zeros=zeros)
+            lp_filter = julius.LowPassFilter(cutoff, zeros=zeros).to(self.device)
+            filtered[i] = lp_filter(self.audio_data[i])
+
         self.audio_data = filtered
         return self

audiotools/core/effects.py

@@ -206,7 +206,7 @@ def equalizer(self, db):
         else:
             db = db.unsqueeze(0)
 
-        weights = (10 ** db).to(self.device).float()
+        weights = (10**db).to(self.device).float()
         fbank = fbank * weights[:, None, None, :]
         eq_audio_data = fbank.sum(-1)
         self.audio_data = eq_audio_data
@@ -252,8 +252,8 @@ def decompose_ir(self):
 
     def measure_drr(self):
         early_response, late_field, _ = self.decompose_ir()
-        num = (early_response ** 2).sum(dim=-1)
-        den = (late_field ** 2).sum(dim=-1)
+        num = (early_response**2).sum(dim=-1)
+        den = (late_field**2).sum(dim=-1)
         drr = 10 * torch.log10(num / den)
         return drr
 
@@ -263,17 +263,17 @@ def solve_alpha(early_response, late_field, wd, target_drr):
         # ----------
         # Apply the good ol' quadratic formula.
 
-        wd_sq = wd ** 2
+        wd_sq = wd**2
         wd_sq_1 = (1 - wd) ** 2
-        e_sq = early_response ** 2
-        l_sq = late_field ** 2
+        e_sq = early_response**2
+        l_sq = late_field**2
         a = (wd_sq * e_sq).sum(dim=-1)
         b = (2 * (1 - wd) * wd * e_sq).sum(dim=-1)
         c = (wd_sq_1 * e_sq).sum(dim=-1) - torch.pow(10, target_drr / 10) * l_sq.sum(
             dim=-1
         )
 
-        expr = ((b ** 2) - 4 * a * c).sqrt()
+        expr = ((b**2) - 4 * a * c).sqrt()
         alpha = torch.maximum(
             (-b - expr) / (2 * a),
             (-b + expr) / (2 * a),

audiotools/core/ffmpeg.py

@@ -2,6 +2,7 @@
 import subprocess
 import tempfile
 
+import ffmpy
 import numpy as np
 import torch
 
@@ -79,10 +80,15 @@ def ffmpeg_resample(self, sample_rate, quiet=True):
     @classmethod
     def load_from_file_with_ffmpeg(cls, audio_path, quiet=True, **kwargs):
         with tempfile.NamedTemporaryFile(suffix=".wav") as f:
-            command = f"ffmpeg -i '{audio_path}' {f.name} -y"
+            global_options = "-y"
             if quiet:
-                command += " -hide_banner -loglevel error"
+                global_options += " -loglevel error"
 
-            subprocess.check_call(shlex.split(command))
+            ff = ffmpy.FFmpeg(
+                inputs={audio_path: None},
+                outputs={f.name: None},
+                global_options=global_options,
+            )
+            ff.run()
             signal = cls(f.name, **kwargs)
         return signal