c-mls-PXR — voice-corrected

What was done to this chain. All 5 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 0.947 before conversion and 0.938 after — it fell by 0.010. Neighbour-to-neighbour the worst pair went 0.938 → 0.910. (The earlier render, with segment 1 left raw, scores 0.771 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, Anger moved +0.416 in the original and +0.339 after conversion — 81 % of the delta retained, which is most of it. On the other named axis, Relief, -0.350 became -0.218.

Quality. Mean predicted overall quality across the segments went 3.19 → 3.45 (+0.27) 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 0.907 before conversion and 0.909 after — it rose by 0.001. Neighbour-to-neighbour the worst pair went 0.904 → 0.909. (The earlier render, with segment 1 left raw, scores 0.788 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, Sexual Lust moved +0.407 in the original and +0.111 after conversion — 27 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Relief, -0.297 became -0.231.

Quality. Mean predicted overall quality across the segments went 3.22 → 3.39 (+0.17) 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 0.889 before conversion and 0.914 after — it rose by 0.025. Neighbour-to-neighbour the worst pair went 0.889 → 0.924. (The earlier render, with segment 1 left raw, scores 0.810 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, Sadness moved +0.537 in the original and +0.022 after conversion — 4 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Triumph, -0.351 became -0.310.

Quality. Mean predicted overall quality across the segments went 3.19 → 3.35 (+0.17) 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 5 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 0.912 before conversion and 0.909 after — it fell by 0.002. Neighbour-to-neighbour the worst pair went 0.914 → 0.886. (The earlier render, with segment 1 left raw, scores 0.800 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, Affection moved +0.354 in the original and +0.259 after conversion — 73 % of the delta retained, which is most of it. On the other named axis, Shame, -0.346 became -0.036.

Quality. Mean predicted overall quality across the segments went 3.22 → 3.43 (+0.21) 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 5 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 0.942 before conversion and 0.940 after — it fell by 0.003. Neighbour-to-neighbour the worst pair went 0.953 → 0.927. (The earlier render, with segment 1 left raw, scores 0.837 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, Jealousy and Envy moved +0.278 in the original and +0.139 after conversion — 50 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Contempt, -0.438 became -0.558.

Quality. Mean predicted overall quality across the segments went 3.13 → 3.31 (+0.19) 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 0.896 before conversion and 0.889 after — it fell by 0.007. Neighbour-to-neighbour the worst pair went 0.902 → 0.909. (The earlier render, with segment 1 left raw, scores 0.744 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, Jealousy and Envy moved +0.279 in the original and +0.209 after conversion — 75 % of the delta retained, which is most of it. On the other named axis, Fear, -0.287 became -0.629.

Quality. Mean predicted overall quality across the segments went 3.01 → 3.36 (+0.35) 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 0.934 before conversion and 0.882 after — it fell by 0.052. Neighbour-to-neighbour the worst pair went 0.942 → 0.889. (The earlier render, with segment 1 left raw, scores 0.832 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, Sourness moved +0.275 in the original and +0.152 after conversion — 55 % of the delta retained. On the other named axis, Emotional Numbness, -0.254 became -0.348.

Quality. Mean predicted overall quality across the segments went 3.22 → 3.30 (+0.08) 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 5 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 0.947 before conversion and 0.922 after — it fell by 0.026. Neighbour-to-neighbour the worst pair went 0.938 → 0.927. (The earlier render, with segment 1 left raw, scores 0.764 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, Anger moved +0.416 in the original and +0.232 after conversion — 56 % of the delta retained. On the other named axis, Relief, -0.350 became -0.329.

Quality. Mean predicted overall quality across the segments went 3.20 → 3.52 (+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 5 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 0.929 before conversion and 0.943 after — it rose by 0.013. Neighbour-to-neighbour the worst pair went 0.925 → 0.943. (The earlier render, with segment 1 left raw, scores 0.725 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, Longing moved +0.295 in the original and +0.119 after conversion — 41 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Anger, -0.320 became -0.171.

Quality. Mean predicted overall quality across the segments went 3.21 → 3.40 (+0.19) 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 0.943 before conversion and 0.938 after — it fell by 0.006. Neighbour-to-neighbour the worst pair went 0.905 → 0.920. (The earlier render, with segment 1 left raw, scores 0.760 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, Thankfulness Gratitude moved +0.418 in the original and +0.394 after conversion — 94 % of the delta retained, which is essentially all of it. On the other named axis, Contemplation, -0.333 became -0.449.

Quality. Mean predicted overall quality across the segments went 3.18 → 3.40 (+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 0.917 before conversion and 0.866 after — it fell by 0.051. Neighbour-to-neighbour the worst pair went 0.924 → 0.832. (The earlier render, with segment 1 left raw, scores 0.802 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, Triumph moved +0.341 in the original and +0.165 after conversion — 48 % of the delta retained, so a meaningful part of the trajectory was flattened. On the other named axis, Emotional Numbness, -0.375 became -0.268.

Quality. Mean predicted overall quality across the segments went 3.22 → 3.33 (+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 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 0.870 before conversion and 0.877 after — it rose by 0.007. Neighbour-to-neighbour the worst pair went 0.855 → 0.852. (The earlier render, with segment 1 left raw, scores 0.617 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, Longing moved +0.288 in the original and +0.160 after conversion — 56 % of the delta retained. On the other named axis, Jealousy and Envy, -0.461 became -0.586.

Quality. Mean predicted overall quality across the segments went 2.65 → 3.23 (+0.58) 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 0.957 before conversion and 0.906 after — it fell by 0.052. Neighbour-to-neighbour the worst pair went 0.950 → 0.872. (The earlier render, with segment 1 left raw, scores 0.795 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, Relief moved +0.313 in the original and +0.210 after conversion — 67 % of the delta retained. On the other named axis, Contemplation, -0.395 became -0.412.

Quality. Mean predicted overall quality across the segments went 3.06 → 3.26 (+0.19) 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 5 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 0.829 before conversion and 0.903 after — it rose by 0.074. Neighbour-to-neighbour the worst pair went 0.864 → 0.901. (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, Anger moved +0.298 in the original and +0.191 after conversion — 64 % of the delta retained. On the other named axis, Helplessness, -0.215 became -0.532.

Quality. Mean predicted overall quality across the segments went 3.27 → 3.43 (+0.17) 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 0.955 before conversion and 0.914 after — it fell by 0.041. Neighbour-to-neighbour the worst pair went 0.945 → 0.908. (The earlier render, with segment 1 left raw, scores 0.796 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, Contentment moved +0.289 in the original and +0.383 after conversion — 132 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Emotional Numbness, -0.475 became -0.171.

Quality. Mean predicted overall quality across the segments went 3.23 → 3.43 (+0.19) 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 0.862 before conversion and 0.878 after — it rose by 0.016. Neighbour-to-neighbour the worst pair went 0.867 → 0.882. (The earlier render, with segment 1 left raw, scores 0.657 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, Concentration moved +0.289 in the original and +0.144 after conversion — 50 % of the delta retained. On the other named axis, Contempt, -0.337 became +0.266.

Quality. Mean predicted overall quality across the segments went 3.04 → 3.39 (+0.36) 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 0.916 before conversion and 0.913 after — it fell by 0.003. Neighbour-to-neighbour the worst pair went 0.916 → 0.913. (The earlier render, with segment 1 left raw, scores 0.712 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, Concentration moved +0.303 in the original and +0.599 after conversion — 197 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Anger, -0.406 became -0.379.

Quality. Mean predicted overall quality across the segments went 3.16 → 3.39 (+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 0.906 before conversion and 0.886 after — it fell by 0.020. Neighbour-to-neighbour the worst pair went 0.906 → 0.886. (The earlier render, with segment 1 left raw, scores 0.777 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, Shame moved +0.272 in the original and +0.349 after conversion — 128 % of the delta retained, i.e. the move came out slightly larger after conversion than before. On the other named axis, Longing, -0.258 became -0.103.

Quality. Mean predicted overall quality across the segments went 3.27 → 3.37 (+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 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 0.945 before conversion and 0.896 after — it fell by 0.049. Neighbour-to-neighbour the worst pair went 0.923 → 0.846. (The earlier render, with segment 1 left raw, scores 0.845 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, Sourness moved +0.288 in the original and +0.276 after conversion — 96 % of the delta retained, which is essentially all of it. On the other named axis, Concentration, -0.263 became -0.291.

Quality. Mean predicted overall quality across the segments went 2.99 → 3.22 (+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 0.904 before conversion and 0.908 after — it rose by 0.003. Neighbour-to-neighbour the worst pair went 0.904 → 0.930. (The earlier render, with segment 1 left raw, scores 0.771 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, Concentration moved +0.289 in the original and +0.198 after conversion — 68 % of the delta retained. On the other named axis, Sourness, -0.341 became -0.332.

Quality. Mean predicted overall quality across the segments went 3.22 → 3.35 (+0.13) 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.