Code Excited Linear Prediction Speech Coding Patents: Who Holds the Core Claims

CELP is a speech-coding architecture that models the vocal tract with linear prediction and encodes the residual using a codebook search, and it underpins a large share of voice codecs used in mobile and VoIP systems. In this dataset, 88.3% of the 3,081 records in scope carry the G10L IPC subclass for speech and audio analysis and synthesis, showing that most patenting activity in this search treats CELP as a core signal-processing technique rather than a peripheral feature. Filings typically combine a CELP encoding layer with a secondary technique, such as transform coding or entropy coding of the residual, as seen in the representative patent US8527265B2.

The assignee ranking behind this search covers 100 companies, with the leading assignee holding 260 of the 3,081 records in scope. The top five combined hold 942 records, 30.6% of all records, and the top ten combined reach 1,511 records, 49.0% of the total, indicating a field where a relatively small number of large telecom and electronics companies hold a substantial, though not overwhelming, share of the filings. The remainder is spread across a long tail of smaller filers, which is typical for a mature signal-processing standard technology.

Filing activity peaked in 2007 at 190 records and has trended lower since. Looking at the most recent period that can be treated as complete, filings fell from 15 in 2021 to 11 in 2024, a -27% change over that three-year span. Because publication lags filing by roughly 18 months, the 2025 and 2026 figures in the raw data are still incomplete and should not be read as evidence of a further, sharper decline.

US8527265B2, assigned to Qualcomm and granted in 2013, covers a specific pipeline: extracting a residual signal from a CELP encoding layer, transforming that residual with a DCT-type transform, and encoding the resulting spectral line positions using a combinatorial lexicographical-index technique. It does not block CELP encoding generally, nor does it block other residual-coding methods such as scalar quantization; its scope is tied specifically to the combinatorial position-coding step applied to a CELP-derived transform spectrum. Teams should check this patent specifically when implementing combinatorial indexing of residual spectral lines, rather than assuming it covers the broader CELP-plus-transform combination.

Among receiving offices in this dataset, the United States leads with 723 filings, followed by the European Patent Office at 595, South Korea at 231, Japan at 187, the WIPO PCT route at 166, and Canada at 162. This spread reflects the global commercial footprint of voice codec standards, where filers typically seek protection across the major telecom equipment and handset markets rather than concentrating in a single jurisdiction.