diff --git a/scripts/loaders/SoundLoader.gd b/scripts/loaders/SoundLoader.gd
index dae5a1f..7e96c36 100644
--- a/scripts/loaders/SoundLoader.gd
+++ b/scripts/loaders/SoundLoader.gd
@@ -189,7 +189,7 @@ const TEX_WIDTH := 2048
 const FILTER_PAD := 3
 func samples_to_texture():
 	var num_samples := INST_NUM + SFX_NUM
-	var header_length := num_samples * 6
+	var header_length := num_samples * 8
 
 	# Create header and unwrapped payload separately first
 	var header_buffer := StreamPeerBuffer.new()
@@ -201,15 +201,16 @@ func samples_to_texture():
 		var nonlooping: bool = loop_begin >= loop_end
 		if nonlooping:
 			loop_begin = loop_end
-			loop_end += 1
-		header_buffer.put_u16(header_length + (payload_buffer.get_position()/2) + FILTER_PAD)  # sample_start
-		header_buffer.put_u16(sample.loop_end + FILTER_PAD)  # sample_length
-		header_buffer.put_u16(sample.loop_begin + FILTER_PAD)  # sample_loop_begin
+			loop_end += 3
+		header_buffer.put_32(header_length + (payload_buffer.get_position()/2) + FILTER_PAD)  # sample_start
+		header_buffer.put_u16(loop_end + FILTER_PAD)  # sample_length
+		header_buffer.put_u16(loop_begin + FILTER_PAD)  # sample_loop_begin
 		header_buffer.put_u16(sample.mix_rate)  # sample_mixrate
 		header_buffer.put_u8(0)  # TODO: attack
 		header_buffer.put_u8(0)  # TODO: decay
 		header_buffer.put_u8(0)  # TODO: sustain
 		header_buffer.put_u8(0)  # TODO: release
+		header_buffer.put_u16(0)  # TODO: unk
 		for i in FILTER_PAD:  # Prepend 3 frames of silence
 			payload_buffer.put_16(0)
 		payload_buffer.put_data(sample.data)  # Copy entire S16LE audio data
@@ -251,7 +252,7 @@ func samples_to_texture():
 	var samples_img = Image.new()
 	samples_img.create_from_data(TEX_WIDTH, rows, false, Image.FORMAT_LA8, data)
 	self.samples_tex = ImageTexture.new()
-	self.samples_tex.create_from_image(samples_img, Texture.FLAG_FILTER)
+	self.samples_tex.create_from_image(samples_img, 0)  #Texture.FLAG_FILTER)
 
 
 var player := AudioStreamPlayer.new()  # Make one for each channel, later
diff --git a/shaders/audio_renderer.gdshader b/shaders/audio_renderer.gdshader
index 02eeef2..04e19fc 100644
--- a/shaders/audio_renderer.gdshader
+++ b/shaders/audio_renderer.gdshader
@@ -98,7 +98,7 @@ vec4 test_writeback(sampler2D tex, vec2 uv) {
 // With the 258 texel header, which uses 3 texels of margin, 255 would be subtracted from the above payload,
 //   leaving 261121 texels for the sample data.
 
-const float HEADER_LENGTH_TEXELS = 6.0;
+const float HEADER_LENGTH_TEXELS = 8.0;
 uniform sampler2D instrument_samples;
 uniform vec2 instrument_samples_size = vec2(2048.0, 128.0);
 uniform float instrument_row_padding = 3.0;  // In case we want to go to cubic filtering
@@ -115,15 +115,21 @@ vec2 get_inst_texel(vec2 xy) {
 	return texture(instrument_samples, xy/instrument_samples_size).xw;
 }
 
+float get_inst_texel_int16(float s) {
+	float x = mod(s, instrument_row_payload) + instrument_row_padding;
+	float y = trunc(s / instrument_row_payload);
+	return unpack_int16(texture(instrument_samples, (vec2(x, y) + 0.5)/instrument_samples_size).xw);
+}
+
 float get_instrument_sample(float instrument_index, float pitch_scale, float t, float t_end) {
 	// t_end is for ADSR purposes
 	float header_offset = instrument_index * HEADER_LENGTH_TEXELS;
-	float sample_start = unpack_uint16(get_inst_texel(vec2(header_offset, 0.0)));  // The true start, after the prepended 3 frames of silence
-	float sample_length = unpack_uint16(get_inst_texel(vec2(header_offset + 1.0, 0.0)));  // 3 frames after the true end, because of how we loop
-	float sample_loop_begin = unpack_uint16(get_inst_texel(vec2(header_offset + 2.0, 0.0)));  // 3 frames after the true loop point
-	float sample_mixrate = unpack_uint16(get_inst_texel(vec2(header_offset + 3.0, 0.0)));
-	vec2 attack_decay = get_inst_texel(vec2(header_offset + 4.0, 0.0));
-	vec2 sustain_release = get_inst_texel(vec2(header_offset + 5.0, 0.0));
+	float sample_start = float(unpack_int32(vec4(get_inst_texel(vec2(header_offset, 0.0)), get_inst_texel(vec2(header_offset + 1.0, 0.0)))));  // The true start, after the prepended 3 frames of silence
+	float sample_length = unpack_uint16(get_inst_texel(vec2(header_offset + 2.0, 0.0)));  // 3 frames after the true end, because of how we loop
+	float sample_loop_begin = unpack_uint16(get_inst_texel(vec2(header_offset + 3.0, 0.0)));  // 3 frames after the true loop point
+	float sample_mixrate = unpack_uint16(get_inst_texel(vec2(header_offset + 4.0, 0.0)));
+	vec2 attack_decay = get_inst_texel(vec2(header_offset + 5.0, 0.0));
+	vec2 sustain_release = get_inst_texel(vec2(header_offset + 6.0, 0.0));
 	// Calculate the point we want to sample in linear space
 	float mixrate = sample_mixrate * pitch_scale;
 	float target_frame = t * mixrate;
@@ -134,10 +140,11 @@ float get_instrument_sample(float instrument_index, float pitch_scale, float t,
 	target_frame -= overshoot_loops*loop_length;
 	// Now we need to identify the sampling point since our frames are spread across multiple rows for GPU reasons
 	// We only sample from texel 4 onwards on a given row - texel 0 is the header, texels 1,2,3 are lead-in for filtering
-	// Note that y should be integral, but x should be continuous, as that's what applies the filtering!
 	target_frame += sample_start;
-	vec2 sample_xy = vec2(instrument_row_padding + mod(target_frame, instrument_row_payload), trunc(target_frame/instrument_row_payload));
-	return rescale_int16(unpack_int16(get_inst_texel(sample_xy)));
+	float a = get_inst_texel_int16(floor(target_frame));
+	float b = get_inst_texel_int16(ceil(target_frame));
+	float mix_amount = fract(target_frame);
+	return rescale_int16(mix(a, b, mix_amount));
 }
 
 const int NUM_CHANNELS = 8;
@@ -196,13 +203,7 @@ vec4 render_song(int smp) {
 		}
 	}
 	// Convert the stereo float audio to S16LE
-	// return vec4(pack_float_to_int16(downmixed_stereo.x), pack_float_to_int16(downmixed_stereo.y));
-	// return vec4(pack_float_to_int16(downmixed_stereo.x), pack_float_to_int16(mod(t, 2.0) - 1.0));
-	vec2 isuv = vec2(mod(float(smp), instrument_samples_size.x), trunc(float(smp)/instrument_samples_size.x))/instrument_samples_size;
-	// float ins = rescale_int16(unpack_int16(texture(instrument_samples, isuv).xw));
-	// return vec4(pack_float_to_int16(ins), pack_float_to_int16(mod(t, 2.0) - 1.0));
-	return vec4(texture(instrument_samples, isuv).xw, pack_float_to_int16(mod(t, 2.0) - 1.0));
-	// return vec4(pack_float_to_int16((t/10.0) - 1.0), pack_float_to_int16(mod(t, 2.0) - 1.0));
+	return vec4(pack_float_to_int16(downmixed_stereo.x), pack_float_to_int16(downmixed_stereo.y));
 }
 
 void fragment() {
diff --git a/test/audio_system.gd b/test/audio_system.gd
index 3d6b063..a671885 100644
--- a/test/audio_system.gd
+++ b/test/audio_system.gd
@@ -138,9 +138,9 @@ func test_rendering() -> void:
 	var midi_events_bytes3 := StreamPeerBuffer.new()
 	var midi_events_bytes4 := StreamPeerBuffer.new()
 	for i in 2048:
-		var t = i * 10.0
+		var t = i * 3.0
 		midi_events_bytes.put_32(t*32000)  # t_start
-		midi_events_bytes2.put_32((t+3.0)*32000)  # t_end
+		midi_events_bytes2.put_32((t+2.75)*32000)  # t_end
 		midi_events_bytes3.put_u8((i%35))  # instrument
 		midi_events_bytes3.put_u8(71)  # pitch_idx
 		# midi_events_bytes.put_float((35 + (i%40)))  # pitch_idx
diff --git a/test/audio_system.tscn b/test/audio_system.tscn
index 2aef8f7..2fa24e3 100644
--- a/test/audio_system.tscn
+++ b/test/audio_system.tscn
@@ -13,7 +13,6 @@ shader_param/instrument_row_payload = 2042.0
 shader_param/reference_note = 71.0
 shader_param/output_mixrate = 32000.0
 shader_param/midi_events_size = Vector2( 2048, 16 )
-shader_param/t_scale = 524.0
 
 [node name="audio_system" type="Node2D"]
 script = ExtResource( 1 )