On Barriers to Archival Audio Processing
This work identifies critical barriers for archives aiming to use speaker recognition for indexing, highlighting biases that affect real-world applications.
The study tested modern language identification and speaker recognition methods on a UNESCO collection of mid-20th century radio recordings, finding that LID systems like Whisper handle accented speech well, but speaker embeddings are fragile and biased by channel, age, and language.
In this study, we leverage a unique UNESCO collection of mid-20th century radio recordings to probe the robustness of modern off-the-shelf language identification (LID) and speaker recognition (SR) methods, especially with respect to the impact of multilingual speakers and cross-age recordings. Our findings suggest that LID systems, such as Whisper, are increasingly adept at handling second-language and accented speech. However, speaker embeddings remain a fragile component of speech processing pipelines that is prone to biases related to the channel, age, and language. Issues which will need to be overcome should archives aim to employ SR methods for speaker indexing.