Reply to thread

Message: [QUOTE="Bombastus, post: 91339, member: 2178"] Illinois law explicitly names a voiceprint as a biometric identifier, but the statute never defines exactly what turns recorded speech into one. That gap matters because an ordinary audio file and biometric data are not automatically the same thing. A voice recording preserves sound. A voiceprint, as courts applying Illinois law have treated the term, involves data capable of identifying a particular speaker from characteristics of that voice. The distinction sounds small, yet it can decide whether the Biometric Information Privacy Act applies at all. That is the narrower issue behind [B][URL='https://goldmidi.com/community/threads/jason-isbell-is-suing-suno-over-ai-identity-use-in-boston.76733/']the Suno voiceprint allegations[/URL][/B]. The complaint points to alleged analysis of vocal features, but saying software measured a voice does not by itself settle whether the resulting data qualifies as a protected biometric identifier. [HEADING=2]Voice characteristics alone do not settle the issue[/HEADING] A 2022 federal ruling in Carpenter v. McDonald's drew a useful line. The court said characteristics such as pitch, volume, duration, accent, and speech pattern, taken individually, are not voiceprints because those traits alone do not uniquely identify someone. It still allowed the claim to proceed because the complaint alleged technology capable of recognizing unique customers across different restaurant visits. That distinction is easy to flatten when discussing AI. Modern audio systems routinely measure acoustic properties for transcription, denoising, language detection, emotion analysis, speaker separation, and other tasks that may have nothing to do with identifying a person. Measuring a formant or estimating pitch therefore does not automatically mean a company created biometric data under Illinois law. The harder question is what the resulting representation can do. Courts have repeatedly focused on whether the collected data is sufficiently tied to a person that it could be used for identification, even when the defendant did not actually identify that person during the disputed transaction. That makes [B]speaker-identification capability[/B] more important than a loose description of the software as voice recognition. Robinson v. Lake Ventures made that practical. Warehouse workers alleged that mandatory onboarding created voice templates later used by a headset system to recognize them while they issued work commands. The court let the claim proceed but said discovery would need to show technology capable of identifying the workers from the stored data, because mere voice-recognition software was insufficient. [HEADING=2]BIPA can apply before identification actually happens[/HEADING] Carpenter also rejected an argument that identification must already have occurred. The court reasoned that collecting a voiceprint capable of identifying someone can trigger BIPA even if the company never uses it to determine who is speaking. The statutory problem can begin at collection rather than at the later moment of successful identification. That position appeared again in Delgado v. Meta Platforms. In 2024, the court found allegations sufficient where Meta's materials allegedly indicated that voice recordings could be used to identify users. The judge treated data unique enough to identify someone as potentially within BIPA even without an allegation that Meta had actually matched the plaintiff to her identity. The case became more useful after discovery. In 2026, the court refused to end the voiceprint claim at summary judgment because evidence created a factual dispute over whether Meta's handling of uploaded audio produced data capable of identifying speakers. The question was not resolved merely by calling the original files recordings. That is a meaningful limit on both sides. A plaintiff cannot simply rename every audio file a voiceprint, but a company cannot necessarily escape by saying it stored only recordings if its processing turns those recordings into an identifying representation. Courts can look past the file label and examine what the system extracts, stores and can infer. [HEADING=2]The technical pipeline can become the legal evidence[/HEADING] For an AI music dispute, the decisive evidence may sit inside the processing pipeline rather than in the final song. Engineers may convert audio into numerical features, embeddings, or other machine-readable representations before training, matching, or generation. Whether those representations qualify under BIPA depends on what they encode and whether they can identify a person. That makes documentation unusually important. Model cards, internal architecture notes, feature-extraction code, patents, retention systems, and expert testing can help show whether a company merely processed sound or built something that functions as an identifier. A marketing label such as voice model or speech model proves very little by itself. The Illinois statute also separates biometric identifiers from biometric information. A voiceprint is expressly listed as an identifier, while biometric information expressly requires information based on an identifier and used to identify someone. In Martell v. X, the court noted that the identifier definition lacks those exact words but still treated identifiability as part of what makes the data an identifier. In an AI-generated music case, words such as pitch, timbre, formants, or breathing patterns do not resolve the biometric issue on their own. The stronger inquiry is whether the system converted those measurements into data that can distinguish and identify the particular speaker, how that data was stored, and what the system was technically capable of doing with it. [/QUOTE]

Name