What the patent claims
Claim 1 sends the same utterance to an on-device recognizer and a cloud service, receives the device result together with features computed from it, and decides, before the cloud result arrives, whether to use the device result or to wait. It then selects the device result.
What we found
SoundHound’s US 9,330,669 (priority June 2012) sends a spoken query to a local and a remote recognizer at the same time, starts a latency timer, and picks a result by recognition score, or takes whichever has answered when the timer runs out.
Nuance’s US 9,953,653 (priority January 2011) runs embedded and remote recognizers on the same audio and lets the device act on the local result without waiting for the remote one.
Two further filings support the combination: AT&T’s US 2015/0120296 (2013) chooses local or network processing from signal strength and latency, and Lenovo’s US 9,620,122 (2011) falls back to a second device when local confidence is low.
None of these appears in the citation record of the ’073 or its family.
Why it matters to the defendant
The ’073 has a 2020 priority date, and the core idea it claims was in the patent literature eight to nine years earlier. The claim’s distinguishing step is the decision rule: a plurality of features from the local result, applied before the cloud answers. SoundHound decides by score or timeout and Nuance acts on the local result early, so the question for a claim chart is how far apart those are from the claimed rule.
The answer in Patent Mechanics
The search as it appears in the app, abridged.
| Reference | Assignee | Priority | Citation record | Fit |
|---|---|---|---|---|
| US 9,330,669 | SoundHound | 2012 | Not cited | Strong |
| US 9,953,653 | Nuance | 2011 | Not cited | Strong |
| US 2015/0120296 | AT&T | 2013 | Not cited | Partial |
| Element | Claim language (abridged) | ’669 | ’653 |
|---|---|---|---|
| 1[a-c] | Same utterance sent to the device and the cloud | ✓ | ✓ |
| 1[d-e] | Device result received with features computed from it | ? | ? |
| 1[f] | Decide before the cloud result whether to use the device result or wait | ? | ✓ |
| 1[g] | Select the device result | ✓ | ✓ |
References
| Reference | Title | Priority | Covers | Gaps |
|---|---|---|---|---|
| US 9,330,669 | Dual mode speech recognition (SoundHound) | 2012-06-21 | Local and remote recognizers in parallel; latency timer; choose by score or on timeout | Decision by score or timeout rather than a set of local-result features |
| US 9,953,653 | Configurable speech recognition using multiple recognizers (Nuance) | 2011-01-07 | Embedded and remote recognizers on the same audio; act on the local result before the remote one | Early action tied to command words and a phone application |
| US 2015/0120296 | Selecting network-based versus embedded speech processing (AT&T) | 2013-10-29 | Chooses local or remote processing from context such as latency | Chooses up front; does not compare results |
Limits of this analysis
- First-pass reading of claims and abstracts; the decision step needs a full claim chart against SoundHound and Nuance.
- Cerence has asserted three other patents in the same case. This article covers only the ’073.
“Not cited” means the reference does not appear in the examiner or applicant citations of the patent or its family in our index. This article is research, not legal advice, and makes no finding that any claim is invalid.
You can do what’s written in this article, and a lot more, today on Patent Mechanics. Try it with a 7-day free trial, no credit card required.