c-snippets-PXR — voice-corrected

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.058 before conversion and 0.670 after — it rose by 0.612. Neighbour-to-neighbour the worst pair went -0.007 → 0.594. (The earlier render, with segment 1 left raw, scores 0.558 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Emotional Numbness moved +0.263 in the original and +0.124 after conversion — 47 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Infatuation, -0.453 became -0.343.

Quality. Mean predicted overall quality across the segments went 2.91 → 3.07 (+0.15) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.433 before conversion and 0.743 after — it rose by 0.310. Neighbour-to-neighbour the worst pair went 0.113 → 0.667. (The earlier render, with segment 1 left raw, scores 0.608 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Infatuation moved +0.371 in the original and +0.669 after conversion — 180 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Hope Enthusiasm Optimism, -0.294 became -0.294.

Quality. Mean predicted overall quality across the segments went 2.72 → 2.92 (+0.20) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.733 before conversion and 0.627 after — it fell by 0.106. Neighbour-to-neighbour the worst pair went 0.733 → 0.712. (The earlier render, with segment 1 left raw, scores 0.383 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Fatigue Exhaustion moved +0.296 in the original and +0.401 after conversion — 136 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Contemplation, -0.272 became -0.168.

Quality. Mean predicted overall quality across the segments went 2.66 → 2.98 (+0.32) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.878 before conversion and 0.844 after — it fell by 0.034. Neighbour-to-neighbour the worst pair went 0.878 → 0.859. (The earlier render, with segment 1 left raw, scores 0.738 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Fatigue Exhaustion moved +0.265 in the original and +0.171 after conversion — 65 % of the delta retained. On the other named axis, Malevolence Malice, -0.261 became -0.333.

Quality. Mean predicted overall quality across the segments went 2.70 → 2.92 (+0.22) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.149 before conversion and 0.288 after — it rose by 0.139. Neighbour-to-neighbour the worst pair went 0.149 → 0.350. (The earlier render, with segment 1 left raw, scores 0.192 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Thankfulness Gratitude moved +0.314 in the original and +0.255 after conversion — 81 % of the delta retained, which is most of it. On the other named axis, Contempt, -0.375 became -0.371.

Quality. Mean predicted overall quality across the segments went 2.85 → 2.89 (+0.03) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.128 before conversion and 0.443 after — it rose by 0.315. Neighbour-to-neighbour the worst pair went 0.064 → 0.416. (The earlier render, with segment 1 left raw, scores 0.237 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Fatigue Exhaustion moved +0.309 in the original and +0.154 after conversion — 50 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Doubt, -0.354 became -0.079.

Quality. Mean predicted overall quality across the segments went 2.21 → 2.68 (+0.48) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.062 before conversion and 0.516 after — it rose by 0.454. Neighbour-to-neighbour the worst pair went 0.269 → 0.474. (The earlier render, with segment 1 left raw, scores 0.428 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Sourness moved +0.583 in the original and +0.639 after conversion — 109 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Infatuation, -0.501 became -0.347.

Quality. Mean predicted overall quality across the segments went 2.54 → 2.68 (+0.15) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.677 before conversion and 0.784 after — it rose by 0.107. Neighbour-to-neighbour the worst pair went 0.737 → 0.828. (The earlier render, with segment 1 left raw, scores 0.571 here.) This chain started 0.50-0.70 — audibly different, a band where the conversion is close to a wash on this measure.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Concentration moved +0.310 in the original and +0.329 after conversion — 106 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Intoxication Altered States of Consciousness, -0.256 became -0.271.

Quality. Mean predicted overall quality across the segments went 2.99 → 3.04 (+0.05) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.077 before conversion and 0.596 after — it rose by 0.519. Neighbour-to-neighbour the worst pair went 0.077 → 0.444. (The earlier render, with segment 1 left raw, scores 0.526 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Emotional Numbness moved +0.399 in the original and +0.123 after conversion — 31 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Concentration, -0.382 became -0.400.

Quality. Mean predicted overall quality across the segments went 2.71 → 2.94 (+0.23) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was -0.137 before conversion and 0.418 after — it rose by 0.555. Neighbour-to-neighbour the worst pair went -0.048 → 0.513. (The earlier render, with segment 1 left raw, scores 0.192 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

The emotional move did not survive. Re-scored end to end, Emotional Numbness moved +0.334 in the original and -0.136 after conversion — it changed direction. On this chain the corrected audio is not an improvement. On the other named axis, Concentration, -0.265 became -0.315.

Quality. Mean predicted overall quality across the segments went 3.05 → 3.16 (+0.11) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.212 before conversion and 0.772 after — it rose by 0.560. Neighbour-to-neighbour the worst pair went 0.212 → 0.755. (The earlier render, with segment 1 left raw, scores 0.606 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Contemplation moved +0.332 in the original and +0.292 after conversion — 88 % of the delta retained, which is most of it. On the other named axis, Affection, -0.314 became -0.883.

Quality. Mean predicted overall quality across the segments went 2.70 → 2.82 (+0.12) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.736 before conversion and 0.758 after — it rose by 0.022. Neighbour-to-neighbour the worst pair went 0.745 → 0.741. (The earlier render, with segment 1 left raw, scores 0.727 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Intoxication Altered States of Consciousness moved +0.239 in the original and +0.266 after conversion — 111 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Sexual Lust, -0.258 became -0.295.

Quality. Mean predicted overall quality across the segments went 2.64 → 2.74 (+0.10) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.446 before conversion and 0.645 after — it rose by 0.199. Neighbour-to-neighbour the worst pair went 0.474 → 0.635. (The earlier render, with segment 1 left raw, scores 0.593 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Pleasure Ecstasy moved +0.360 in the original and +0.697 after conversion — 193 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Sexual Lust, -0.336 became -0.234.

Quality. Mean predicted overall quality across the segments went 2.86 → 3.08 (+0.22) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.606 before conversion and 0.770 after — it rose by 0.164. Neighbour-to-neighbour the worst pair went 0.677 → 0.771. (The earlier render, with segment 1 left raw, scores 0.571 here.) This chain started 0.50-0.70 — audibly different, a band where the conversion is close to a wash on this measure.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Contemplation moved +0.299 in the original and +0.290 after conversion — 97 % of the delta retained, which is essentially all of it. On the other named axis, Concentration, -0.376 became -0.429.

Quality. Mean predicted overall quality across the segments went 2.89 → 3.03 (+0.14) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was -0.206 before conversion and 0.256 after — it rose by 0.462. Neighbour-to-neighbour the worst pair went -0.089 → 0.256. (The earlier render, with segment 1 left raw, scores 0.117 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Infatuation moved +0.494 in the original and +0.137 after conversion — 28 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Intoxication Altered States of Consciousness, -0.426 became -0.498.

Quality. Mean predicted overall quality across the segments went 2.69 → 2.75 (+0.07) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was -0.123 before conversion and 0.503 after — it rose by 0.626. Neighbour-to-neighbour the worst pair went -0.045 → 0.455. (The earlier render, with segment 1 left raw, scores 0.473 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Fear moved +0.400 in the original and +0.576 after conversion — 144 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Emotional Numbness, -0.394 became -0.605.

Quality. Mean predicted overall quality across the segments went 2.60 → 2.75 (+0.15) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.705 before conversion and 0.811 after — it rose by 0.107. Neighbour-to-neighbour the worst pair went 0.705 → 0.811. (The earlier render, with segment 1 left raw, scores 0.720 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Infatuation moved +0.278 in the original and +0.182 after conversion — 65 % of the delta retained. On the other named axis, Astonishment Surprise, -0.407 became -0.591.

Quality. Mean predicted overall quality across the segments went 2.85 → 2.90 (+0.05) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.018 before conversion and 0.330 after — it rose by 0.313. Neighbour-to-neighbour the worst pair went -0.006 → 0.415. (The earlier render, with segment 1 left raw, scores 0.174 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Intoxication Altered States of Consciousness moved +0.404 in the original and +0.207 after conversion — 51 % of the delta retained. On the other named axis, Fatigue Exhaustion, -0.328 became -0.417.

Quality. Mean predicted overall quality across the segments went 2.63 → 2.75 (+0.12) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 4 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.072 before conversion and 0.329 after — it rose by 0.257. Neighbour-to-neighbour the worst pair went 0.065 → 0.286. (The earlier render, with segment 1 left raw, scores 0.066 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Emotional Numbness moved +0.297 in the original and +0.152 after conversion — 51 % of the delta retained. On the other named axis, Impatience and Irritability, -0.329 became -0.314.

Quality. Mean predicted overall quality across the segments went 2.60 → 2.75 (+0.15) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.

What was done to this chain. All 3 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.

Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.033 before conversion and 0.341 after — it rose by 0.307. Neighbour-to-neighbour the worst pair went 0.033 → 0.341. (The earlier render, with segment 1 left raw, scores 0.199 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.

Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Astonishment Surprise moved +0.347 in the original and +0.582 after conversion — 167 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Pride, -0.354 became -0.053.

Quality. Mean predicted overall quality across the segments went 2.66 → 2.88 (+0.23) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.