Across the 87 records in this dataset, one assignee leads with 11 records, roughly triple the count held by the fifth-ranked assignee at 4. The top 5 assignees combined account for 39.1% of all records, and the top 10 account for 59.8%, which means well over a third of the field sits with a long tail of smaller filers rather than a small handful of dominant firms. This pattern suggests the field has an established leader but is still open enough for new entrants to carve out claim space.

The trend rose from 0 records in 2017 to a peak of 9 in 2024, after sitting flat at 1 record at the 2022 midpoint. Filing counts drop off after 2024 into the partial 2026 year, and none of the most recently active assignees logged a filing in the latest tracked year. Because publication typically lags filing by around 18 months, the most recent years understate real activity, but the overall shape points to a field that has already passed its filing peak rather than one still accelerating.

The dominant classification is G10L, speech and audio analysis and synthesis, which appears on 97.7% of the 87 records in scope. G06F (electric digital data processing) appears on 12.6%, G06N (AI-based computing models) on 4.6%, H04M (telephonic communication) on 3.4%, and G11B (information storage) on 1.1%. Because a single record can carry multiple IPC classes, these shares add up to more than 100%, and the low overlap with G06N and H04M suggests claims explicitly framed around AI-model architecture or telephony integration are still comparatively rare.

US8407054B2, assigned to NEC Corporation and dated 2013-03-26, claims a speech synthesis device that selects a central speech segment, generates prosody information from it, then selects non-central segments based on both the central segment and that prosody information, before generating a final synthesized waveform. The specific sequencing — prosody generation gated on central-segment selection before non-central segments are chosen — is the part later filers need to check against, since copying that exact order without variation risks overlap. It is a useful reference point for anyone drafting segment-selection or prosody-generation claims in this space.

The clearest gaps sit in areas that overlap only lightly with the dominant G10L cluster: voice cloning consent safeguards, real-time-factor optimisation for on-device vocoders, multilingual prosody transfer, and synthesis claims integrated with telephony systems (H04M) or explicit AI-model architectures (G06N). These branches show meaningfully lower representation than core waveform and prosody synthesis claims, which is where most of the 87 records concentrate. A first claim in one of these areas would need to tie the technical mechanism (for example, a specific consent-verification step in a cloning pipeline) to a measurable output like naturalness MOS or real-time factor, rather than restating the general synthesis pipeline already well covered by existing filings.