Estimating speaker distance from a single micro- phone is challenging in reverberant environments, where acoustic cues depend on the room. In this work, we analyze the impact of different room impulse response (RIR) components on distance estimation. A synthetic dataset is generated using shoebox rooms with random geometries, source–microphone positions, and wall absorption coefficients. For each configuration, the RIR is divided into four variants: full response, direct path only, response without late reverberation, and response without early reflections. The results show very similar errors for all variants, in both the fully calibrated and time-calibrated scenarios. This suggests that, in the considered setup, the model does not strongly depend on the isolated reverberant components, but may exploit shared cues such as propagation delay.
Stimare la distanza di un parlatore con un singolo microfono `e difficile in ambienti riverberanti, dove gli indizi acustici dipendono dalla stanza. In questo lavoro analizziamo l’impatto di diverse componenti della risposta impulsiva della stanza (RIR) sulla stima della distanza. Viene generato un dataset sintetico di stanze shoebox con geometrie, posizioni sorgente– microfono e assorbimenti casuali. Per ogni configurazione, la RIR viene suddivisa in quattro varianti: risposta completa, solo percorso diretto, risposta senza riverberazione tardiva e risposta senza riflessioni precoci. I risultati mostrano errori molto simili per tutte le varianti, sia nello scenario fully calibrated sia in quello time-calibrated. Questo suggerisce che, nel setup considerato, il modello non dipenda fortemente dalle componenti riverberanti isolate, ma possa sfruttare indizi condivisi, come il ritardo di propagazione.
Analyzing RIR Impact for Single-Channel Speaker Distance Estimation
AGHAMOHAMMADI, SHIVA
2025/2026
Abstract
Estimating speaker distance from a single micro- phone is challenging in reverberant environments, where acoustic cues depend on the room. In this work, we analyze the impact of different room impulse response (RIR) components on distance estimation. A synthetic dataset is generated using shoebox rooms with random geometries, source–microphone positions, and wall absorption coefficients. For each configuration, the RIR is divided into four variants: full response, direct path only, response without late reverberation, and response without early reflections. The results show very similar errors for all variants, in both the fully calibrated and time-calibrated scenarios. This suggests that, in the considered setup, the model does not strongly depend on the isolated reverberant components, but may exploit shared cues such as propagation delay.| File | Dimensione | Formato | |
|---|---|---|---|
|
final_presentation_shiva.pdf
Accesso riservato
Dimensione
697.99 kB
Formato
Adobe PDF
|
697.99 kB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/111489