diff --git a/audio.py b/audio.py
index 10b233e..5346680 100644
--- a/audio.py
+++ b/audio.py
@@ -1,34 +1,87 @@
 import numpy as np
+import audioread
 import librosa
-import soundfile as sf
 
 import opcodes
 import video
 
+TICKS = {
+    4: opcodes.Tick4,
+    6: opcodes.Tick6,
+    8: opcodes.Tick8,
+    10: opcodes.Tick10,
+    12: opcodes.Tick12,
+    14: opcodes.Tick14,
+    16: opcodes.Tick16,
+    18: opcodes.Tick18,
+    20: opcodes.Tick20,
+    22: opcodes.Tick22,
+    24: opcodes.Tick24,
+    26: opcodes.Tick26,
+    28: opcodes.Tick28,
+    30: opcodes.Tick30,
+    32: opcodes.Tick32,
+    34: opcodes.Tick34,
+    36: opcodes.Tick36,
+    38: opcodes.Tick38,
+    40: opcodes.Tick40,
+    42: opcodes.Tick42,
+    44: opcodes.Tick44,
+    46: opcodes.Tick46,
+    48: opcodes.Tick48,
+    50: opcodes.Tick50,
+    52: opcodes.Tick52,
+    54: opcodes.Tick54,
+    56: opcodes.Tick56,
+    58: opcodes.Tick58,
+    60: opcodes.Tick60,
+    62: opcodes.Tick62,
+    64: opcodes.Tick64,
+    66: opcodes.Tick66,
+}
+
 
 class Audio:
     def encode_audio(self, audio):
         for a in audio:
-            yield opcodes.Tick(a)
-            yield opcodes.Tick(50-a)
+            a = max(-30, min(a * 2, 32)) + 34
+            yield TICKS[a]()
+
 
 def main():
-    filename = librosa.util.example_audio_file()
-
-    data, samplerate = sf.read(filename, dtype='float32')
-    data = data.T
-    a = librosa.resample(data, samplerate, 20000).flatten()
-
-    a = librosa.util.normalize(a)
-    a = (a * 10 + 25).astype(np.int)
+    filename = "Computer Chronicles - 06x05 - The Apple II.mp4"
 
     s = video.Video(frame_rate=None)
     au = Audio()
 
-    with open("out.bin", "wb") as out:
-        out.write(bytes(s.emit_stream(au.encode_audio(a))))
-        out.write(bytes(s.done()))
+    with audioread.audio_open(filename) as f:
+        with open("out.bin", "wb") as out:
+            for buf in f.read_data(128 * 1024):
+                print(f.channels, f.samplerate, f.duration)
+
+                data = np.frombuffer(buf, dtype='int16').astype(
+                    'float32').reshape((f.channels, -1), order='F')
+
+                a = librosa.core.to_mono(data)
+                a = librosa.resample(a, f.samplerate, 14000).flatten()
+
+                # Normalize to 95%ile
+                # norm = max(
+                #    abs(np.percentile(a, 5, axis=0)),
+                #    abs(np.percentile(a, 95, axis=0))
+                # )
+                # print(min(a),max(a))
+                # print(norm)
+
+                # XXX how to estimate normalization without reading whole file?
+                norm = 12000
+
+                a /= norm  # librosa.util.normalize(a)
+                a = (a * 32).astype(np.int)
+
+                out.write(bytes(s.emit_stream(au.encode_audio(a))))
+    out.write(bytes(s.done()))
 
 
 if __name__ == "__main__":
-    main()
\ No newline at end of file
+    main()