Speech BCI (Cortical Surface, Avatar Speech) by Echo Neurotechnologies: status and timeline

Speech BCI (Cortical Surface, Avatar Speech)

Cortical surface electrodes decode speech neural signals for paralysed patients. Outputs: text, speech, avatar animation.

PilotStage 3 of 5

Clinical trials ongoing; 78 wpm text, 25% error rate. Neurotech Leaders 2025 award.

Updated 30 Jan 2025Checked 10 Oct0 updates this week

Milestones

Next · Expanded clinical trials with additional patients

Echo Neurotechnologies founded2023Complete.
Series A funding led by Andreessen HorowitzJan 2025Complete.
Clinical trial demonstrates 78 wpm text decoding2025Complete.
Most Promising Startup award at Neurotech Leaders Forum 20252025Complete.
Expanded clinical trials with additional patientsTarget 2026Current milestone.
Path to FDA approvalTarget 2027Not yet reached.

Most important updates

Upcoming

2026Expanded clinical trials with additional patients (next)
2026FDA breakthrough device designation (if awarded)
2027Path to FDA approval

Current obstacles

Expanding beyond trained vocabularyCurrent system works best within a trained word set; decoding spontaneous, unrestricted speech with unlimited vocabulary remains a major challenge.
Clinical translation and surgical logisticsMoving from research trials at UCSF to broader clinical availability requires FDA approval, surgeon training and manufacturing scale-up.
Electrode degradation over timeECoG signal quality can degrade over months as glial scar tissue forms; maintaining stable long-term performance is a major engineering challenge.

Physics limits

Cortical plasticity and signal variabilityBrain signals change over time due to learning, fatigue and neural plasticity; decoders must adapt continuously to maintain performance.
Spatial coverage of electrode arrayA single electrode array samples a limited cortical area; larger coverage would require larger implants or multiple implants, increasing surgical complexity and risk.
Latency in synthesis and feedbackConverting neural signals to speech takes time; minimising latency below 100 milliseconds is needed for natural conversational flow.

How it works

3 parts

Recording electrodes

Cortical surface array (ECoG)

High-density grids on speech cortex. Better SNR than deep electrodes, less invasive.

Neural decoding

Machine learning models

Algorithms trained on neural recordings and the user's intended speech decode real-time neural signals to predict spoken words, generating text output.

Speech synthesis

Avatar and audio output

Decoded words become speech audio and avatar facial animation for natural communication.

Spec sheet

Spec	Speech BCI (Cortical Surface, Avatar Speech)
Text output speed	78 words per minute^R (reported)
Median word error rate	25%^R (reported)
Time to high performance	Less than 2 weeks^R (reported)

^R reported by the company

Papers & demos

1 item

2023paper
High-performance neuroprosthesis for speech decoding and avatar control using ECoG (opens pmc.ncbi.nlm.nih.gov)Foundational study demonstrating real-time speech decoding from cortical surface recordings with 78 wpm text output.

Update log

3 updates

About Echo Neurotechnologies

The team behind Speech BCI (Cortical Surface, Avatar Speech)

Echo Neurotechnologies

Speech BCI, cortical surface, avatar speech synthesis

Echo develops BCIs for speech restoration in paralysed patients. Founded 2023, San Francisco; led by Edward Chang (UCSF). a16z-backed Series A: $50M (Jan 2025).

ECoG array decodes speech signals; synthesises speech via avatar.
Achieved 78 wpm text, 25% error rate in trials with <2 weeks training.
Neurotech Leaders 2025 award; a16z $50M Series A (Jan 2025).

echo.bio (opens echo.bio)Company profile →

Founded: 20233 yrs
Headquarters: United States
Status: Private
Valuation: private
Raised: $50M1 round
Last round: Series A · $50MJan 2025
Works in: NeurotechBrain implants
Coverage: 1 program · 3 updateslatest 30 Jan 2025checked 10 Oct
Lead investors: Andreessen Horowitz
Partners: University of California, San Francisco (opens ucsf.edu)
Investors: Andreessen Horowitz (opens a16z.com)
People: Edward Chang (opens changlab.ucsf.edu)Founder and Chief Science Officer

Spec

Speech BCI (Cortical Surface, Avatar Speech)

Text output speed

78 words per minute^R (reported)

Median word error rate

25%^R (reported)

Time to high performance

Less than 2 weeks^R (reported)

Speech BCI (Cortical Surface, Avatar Speech)

Milestones

Most important updates

Upcoming

Current obstacles

Physics limits

How it works

Cortical surface array (ECoG)

Machine learning models

Avatar and audio output

Spec sheet

Papers & demos

Update log

Thu 30 Jan 2025

2025

Aug 2023

About Echo Neurotechnologies

Current obstacles

Physics limits

How it works

Cortical surface array (ECoG)

Machine learning models

Avatar and audio output

Spec sheet

Papers & demos

Update log

Thu 30 Jan 2025

2025

Aug 2023

About Echo Neurotechnologies