Remove melodic note choke, fix banjo to 16th notes

The melodic renderer was choking all signal at each new note,
killing strum/hold resonance. Removed — notes now naturally
overlap and ring out. Banjo lick now uses 16th notes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

docs/generate_audio.py

@@ -667,10 +667,10 @@ def gen_synth_banjo():
     # Strum into a picking lick
     p.strum("G", Duration.WHOLE, velocity=80)
     p.strum("C", Duration.WHOLE, velocity=78)
-    # Bluegrass lick
+    # Bluegrass lick — 16th note picking
     for n in ["G4", "B4", "D5", "G5", "D5", "B4", "A4", "G4",
               "D4", "G4", "B4", "D5", "B4", "G4", "D4", "G4"]:
-        p.add(n, Duration.EIGHTH, velocity=82)
+        p.add(n, Duration.SIXTEENTH, velocity=82)
     render("synth_banjo", score)
 
 def gen_synth_mandolin():

pytheory/play.py

@@ -4607,12 +4607,6 @@ def _render_notes_to_buf(notes, buf, samples_per_beat, total_samples,
                     vel_cutoff = vel_to_filter * vel_scale + 1000
                     mixed = _apply_lowpass(mixed, vel_cutoff, q=filter_q)
                 end = min(start + len(mixed), total_samples)
-                # Choke: fade out any existing signal at this point
-                # so new notes don't pile up on previous tails
-                choke_len = min(int(SAMPLE_RATE * 0.003), start)
-                if choke_len > 0:
-                    fade = numpy.linspace(1.0, 0.0, choke_len).astype(numpy.float32)
-                    buf[start - choke_len:start] *= fade
                 buf[start:end] += mixed[:end - start] * volume * vel_scale
                 # Spread detuned oscillators into stereo L/R
                 if detune_up is not None and stereo_buf is not None: