diff --git a/.gitignore b/.gitignore
index b948985..492ee29 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,10 @@
 *.swp
 *.pyc
+gongs_wav
+CLIPS.json
+DRONES.json
+GONGS.json
+MUSIC.json
+PATHS.json
+VOCALS.json
+
diff --git a/gongs_wav.py b/gongs_wav.py
new file mode 100755
index 0000000..58aaa29
--- /dev/null
+++ b/gongs_wav.py
@@ -0,0 +1,12 @@
+#!/usr/bin/python3
+from glob import glob
+import subprocess
+import os
+
+for f in glob('gongs/*/*.mp3'):
+    t = f.replace('gongs/', 'gongs_wav/').replace('.mp3', '.wav')
+    print(f, t)
+    folder = os.path.dirname(t)
+    if not os.path.exists(folder):
+        os.makedirs(folder)
+    subprocess.call(['ffmpeg', '-i', f, t])
diff --git a/pi.py b/pi.py
index 7b3ebd1..491a633 100644
--- a/pi.py
+++ b/pi.py
@@ -6,6 +6,7 @@ class random(object):
     PI = str(mp.pi).replace('.', '')
 
     def __init__(self, offset=0):
+        self.position = offset
         self.numbers = list(map(int, self.PI[offset:]))
 
     def __call__(self):
@@ -14,5 +15,6 @@ class random(object):
             mp.dps += 1000
             self.PI = str(mp.pi).replace('.', '')
             self.numbers = list(map(int, self.PI[offset:]))
+        self.position += 1
         return self.numbers.pop(0)
 
diff --git a/render.py b/render.py
index 19b05e8..726c5a9 100755
--- a/render.py
+++ b/render.py
@@ -419,6 +419,10 @@ def sequence(seq, letter):
             tduration = sum([c['duration'] for c in result[track]])
             if not abs(tduration - duration) < 0.000001:
                 raise Exception('invalid duration on track: %s %s vs %s %s' % (track, tduration, duration, result[track]))
+    result['gongs'] = {
+        'tracks': 100 + seq() * 23,
+        'offset': seq.position,
+    }
     return result
 
 
diff --git a/render_gongs.py b/render_gongs.py
new file mode 100755
index 0000000..755e88e
--- /dev/null
+++ b/render_gongs.py
@@ -0,0 +1,132 @@
+#!/usr/bin/python3
+import os
+import sys
+import json
+import subprocess
+from collections import defaultdict
+import string
+from glob import glob
+
+import nmt
+import numpy as np
+
+from pi import random
+from keywords import KEYWORDS
+import ox
+import ox.web.auth
+
+
+if not os.path.exists('GONGS.json'):
+    GONGS = defaultdict(list)
+    prefix = 'gongs_wav'
+    wavs = glob('%s/*/*.wav' % prefix)
+    for path in sorted(wavs):
+        name = os.path.basename(path).split('.')[0]
+        GONGS[name] = {
+            'path': path,
+            'duration': ox.avinfo(path)['duration']
+        }
+    with open('GONGS.json', 'w') as fd:
+        json.dump(GONGS, fd, indent=2, sort_keys=True)
+else:
+    GONGS = json.load(open('GONGS.json'))
+
+
+def random_choice(seq, items):
+    n = n_ = len(items) - 1
+    #print('len', n)
+    if n == 0:
+        return items[0]
+    r = seq()
+    base = 10
+    while n > 10:
+        n /= 10
+        #print(r)
+        r += seq()
+        base += 10
+    r = int(n_ * r / base)
+    #print('result', r, items)
+    return items[r]
+
+
+def splitint(number, by):
+    div = int(number/by)
+    mod = number % by
+    return [div + 1 if i > (by - 1 - mod) else div for i in range(by)]
+
+
+def add_blank(track, d):
+    if track and track[-1].get('blank'):
+        track[-1]['duration'] += d
+    else:
+        blank = {'blank': True, 'duration': d}
+        track.append(blank)
+    return d
+
+
+def get_gongs(seq, result, duration, tracks=42):
+    for tn in range(tracks):
+        track = 'gongs%d' % tn
+        if track not in result:
+            result[track] = []
+
+        position = 0
+        while position < duration:
+            n1 = seq()
+            n2 = seq()
+            clip = GONGS['%d_%d' % (n1, n2)]
+            if position + clip['duration'] < duration:
+                position += clip['duration']
+                result[track].append(clip)
+            else:
+                c = clip.copy()
+                c['duration'] = duration - position
+                result[track].append(c)
+                position += c['duration']
+        if position < duration:
+            position += add_blank(result[track], duration - position)
+
+    for track in result:
+        if result[track]:
+            tduration = sum([c['duration'] for c in result[track]])
+            if not abs(tduration - duration) < 0.000001:
+                raise Exception('invalid duration on track: %s %s vs %s %s' % (track, tduration, duration, result[track]))
+    return result
+
+
+def mix_gongs(data, output):
+    mixed = None
+    for n, clips in enumerate(data.values()):
+        track = [nmt.sound2np(c['path']) for c in clips]
+        track = np.vstack(track)
+        if mixed is None:
+            mixed = track.astype(np.int)
+        else:
+            limit = min(len(track), len(mixed))
+            mixed = mixed[:limit]
+            track = track[:limit]
+            mixed += track.astype(np.int)
+        print(n)
+
+    d = np.sqrt(len(data))
+    while \
+            np.max((mixed/d).astype(np.int)) != np.max((mixed/d).astype(np.int16)) \
+            or np.min((mixed/d).astype(np.int)) != np.min((mixed/d).astype(np.int16)):
+        d = d + 2
+    mixed = (mixed/d).astype(np.int16)
+    nmt.np2sound(mixed, output)
+
+
+if __name__ == '__main__':
+    n = int(sys.argv[1])
+    seq = random(n)
+    duration = float(sys.argv[2]) + 10
+    tracks = int(sys.argv[3])
+    if len(sys.argv) > 1:
+        tracks = int(sys.argv[1])
+    else:
+        tracks = 42
+    data = get_gongs(seq, {}, duration, tracks)
+    mix_gongs(data, output)
+    with open(output + '.json', 'w') as fd:
+        json.dump(result, fd, indent=4, sort_keys=True)
diff --git a/render_mlt.py b/render_mlt.py
index 1f510a4..d182675 100755
--- a/render_mlt.py
+++ b/render_mlt.py
@@ -50,8 +50,7 @@ def add_clip(playlist, clip, in_, duration):
         tractor.plant_filter(volume)
     playlist.append(tractor)
 
-def add_audio_clip(playlist, file_, duration):
-    in_ = 0
+def add_audio_clip(playlist, file_, duration, in_=0):
     if not isinstance(file_, str):
         file_ = file_.encode('utf-8')
     clip = mlt.Producer(profile, file_)
@@ -109,6 +108,7 @@ target_vocals = source.replace('.json', '.vocals.xml')
 target_music = source.replace('.json', '.music.xml')
 target_drones = source.replace('.json', '.drones.xml')
 target_source = source.replace('.json', '.source.xml')
+gongs_wav = source.replace('.json', '.gongs.wav')
 
 with open(source) as fd:
     data = json.load(fd)
@@ -119,12 +119,14 @@ music = mlt.Playlist()
 vocals = mlt.Playlist()
 drones0 = mlt.Playlist()
 drones1 = mlt.Playlist()
+gongs = mlt.Playlist()
 
 # hide Set to 1 to hide the video (make it an audio-only track),
 # 2 to hide the audio (make it a video-only track),
 # or 3 to hide audio and video (hidden track).
 drones0.set("hide", 1)
 drones1.set("hide", 1)
+gongs.set("hide", 1)
 vocals.set("hide", 1)
 music.set("hide", 1)
 
@@ -194,8 +196,25 @@ save_xml(drones, target_drones)
 
 save_xml(music, target_music)
 
+# render gongs
+subprocess.call([
+    # offset in pi, duration, number of tracks, target
+    './render_gongs.py',
+    str(data['gongs']['offset']),
+    str(duration),
+    str(data['gongs']['tracks']),
+    gongs_wav
+])
+# load gongs
+add_audio_clip(gongs, gongs_wav, int(duration * fps), int(5*fps))
+
+
 # mix drones + music
-mtractor = mix_audio_tracks(drones, music, 0.3)
+#mtractor = mix_audio_tracks(drones, music, 0.3)
+
+# mix gongs + music
+mtractor = mix_audio_tracks(gongs, music, 0.3)
+
 norm = mlt.Filter(profile, "volume")
 # lower volume
 norm.set("gain", "-12dB")