## Line Graphs: MUSHRA Score Comparison Across Codecs and Content Types ### Overview Three line graphs compare MUSHRA scores (0-90 scale) for three codecs (Reference, Proposed, EnCodec) across three content types (speech, music, environment) at varying bitrates (2-12 kbps). Each graph shows performance trends with shaded confidence intervals. ### Components/Axes - **X-axis**: Bitrate (kbps), labeled "Bitrate (kbps)", range 2-12 kbps - **Y-axis**: MUSHRA score (0-90), labeled "MUSHRA score", range 0-90 - **Legends**: Positioned right-aligned, color-coded: - Blue circle: Reference (speech/music/env) - Orange square: Proposed (speech/music/env) - Green diamond: EnCodec (speech/music/env) - **Shading**: Gray bands around lines indicate 95% confidence intervals ### Detailed Analysis #### Speech Content - **Reference**: Flat line at 80 (no variation) - **Proposed**: Starts at ~50 (2 kbps), rises to ~75 (8 kbps), peaks at ~78 (12 kbps) - **EnCodec**: Starts at ~15 (2 kbps), rises to ~30 (8 kbps), ends at ~35 (12 kbps) - **Key Points**: Proposed crosses 70 at ~6 kbps; EnCodec crosses 30 at ~6 kbps #### Music Content - **Reference**: Flat line at 80 - **Proposed**: Starts at ~45 (2 kbps), rises to ~65 (8 kbps), peaks at ~68 (12 kbps) - **EnCodec**: Starts at ~20 (2 kbps), rises to ~35 (8 kbps), ends at ~40 (12 kbps) - **Key Points**: Proposed crosses 60 at ~6 kbps; EnCodec crosses 35 at ~6 kbps #### Environment Content - **Reference**: Flat line at 80 - **Proposed**: Starts at ~35 (2 kbps), rises to ~60 (8 kbps), peaks at ~65 (12 kbps) - **EnCodec**: Starts at ~25 (2 kbps), rises to ~40 (8 kbps), ends at ~45 (12 kbps) - **Key Points**: Proposed crosses 60 at ~8 kbps; EnCodec crosses 40 at ~8 kbps ### Key Observations 1. **Performance Gaps**: All codecs underperform Reference (80) across all content types 2. **Bitrate Dependency**: Proposed and EnCodec scores improve with higher bitrates 3. **Codec Comparison**: Proposed consistently outperforms EnCodec at equivalent bitrates 4. **Confidence Intervals**: Shading shows ~±5 score variability for Proposed/EnCodec 5. **Speech Advantage**: Proposed achieves highest absolute scores in speech content ### Interpretation The data demonstrates codec performance tradeoffs: - **Proposed codec** offers best quality/bitrate efficiency but requires higher bitrates to approach Reference quality - **EnCodec** shows lower performance but maintains more consistent gains with bitrate increases - **Content type sensitivity**: Speech benefits most from bitrate increases, while environment shows least improvement - **Reference stability**: Constant 80 score suggests idealized benchmark conditions - **Practical implications**: Proposed codec may be optimal for speech applications where bitrate flexibility exists, while EnCodec might suit constrained environments despite lower ceiling performance