Search “Speech”

DingTalk A1 Doukou Doctor Edition combines voice recording with medical evidence-based capabilities

On October 30, 2026, DingTalk A1 Doukou Doctor Edition was released, combining A1's voice recording and AI meeting notes capabilities with Doukou's medical evidence-based capabilities. According to a China.com Technology report, after doctors finish organizing case discussions or lecture content, they can also query literature, guidelines, and expert consensus around unresolved medical questions. A gynecology department director at a tertiary hospital in Hangzhou reported that on surgery days she can first record special situations and judgments, then organize them after surgery, while for complex outpatient visits she uses Doukou Doctor to look up literature and guideline links. The report states that at this stage it is more like a portable professional assistant capable of recording, distilling, and searching literature than a reconstruction of hospital IT systems.

DingTalk A1 Doukou Doctor Edition combines voice recording and medical evidence-based capabilities

On October 30, 2026, DingTalk A1 Doukou Doctor Edition was released, combining A1's voice recording and AI meeting notes capabilities with Doukou's medical evidence-based capabilities. According to a China.com Technology report, after doctors finish organizing case discussions or lecture content, they can also query literature, guidelines, and expert consensus around unresolved medical questions. A gynecology department director at a tertiary hospital in Hangzhou reported that on surgery days she can first record special situations and judgments, then organize them after surgery, while for complex outpatient visits she uses Doukou Doctor to look up literature and guideline links. The report states that at this stage it is more like a portable professional assistant capable of recording, distilling, and searching literature than a reconstruction of hospital IT systems.

View story:DingTalk A1 Doukou Doctor Edition combines voice recording with medical evidence-based capabilities

Cartesia launches the Managed Agents voice agent platform

Cartesia has launched Managed Agents, an end-to-end voice agent platform for enterprise call scenarios. According to information released by Cartesia's product and research teams, the platform ranks first among 30 agents on Artificial Analysis's Speech Agent Arena with a preference Elo of 1115; on ServiceNow's EVA-Bench it achieved an accuracy of 87.4% and an experience score of 90.7%, both the highest. The platform co-designs the voice model and harness, offering capabilities including 44 languages, tool calling, knowledge bases, and outbound calling batches, and the company says it has already supported tens of millions of conversations in finance, retail, hospitality, and healthcare scenarios. Rohan Paul's October 30 report likewise noted that the platform ranks first on Speech Agent Arena's blind human-preference testing and on EVA-Bench, and said its H-Nets achieve a 3–5x efficiency improvement in controlled scaling experiments.

ElevenLabs partners with adELA to provide free voice restoration to ALS patients in Spain

ElevenLabs announced a partnership with the Spanish ALS association adELA to give ALS patients diagnosed in Spain free access to its voice restoration and preservation technology. This is the first project of the ElevenLabs Impact Program in Spain, and Spain is also the first country to offer this technology through a nationwide patient association. Patients whose voices have not yet changed noticeably can record new samples; those whose voices have already changed can upload old recordings to generate their own digital voice. It is estimated that there are about 4,000–4,500 ALS patients in Spain, many of whom will eventually lose the ability to speak.

Block releases full-duplex speech model Dessa

Block's AI research lab has released the full-duplex speech model Dessa, opening a waitlist as a research preview. According to MarkTechPost, the model was trained only on human voice conversation audio, without using text or pretrained models, and can run locally on a MacBook Pro M4 Max or higher configuration, faster than real time; training used one NVIDIA GB200 NVL72 (72 GPUs). No benchmark scores, parameter count, weights, paper, or API have been released so far.

Databricks launches voice Agent solution and open-sources DIVA blueprint

On October 29, 2026, Databricks launched a voice agent solution and open-sourced the DIVA blueprint as the starting point for the next scenario. The solution emphasizes that the value of voice agents lies not in being able to converse, but in combining context, ontology, and governed data to understand callers in real time and execute rules. Its prototype uses LiveKit to carry calls, Deepgram for speech-to-text and default voice, and ElevenLabs to provide expressive voice, and can hand calls over to a human. The agent runs where the customer's data resides, connecting to order, loyalty, and CRM systems through read-only queries. Currently, the solution uses the DIVA blueprint as the starting point for subsequent scenarios.

ElevenLabs partners with adELA to provide free voice restoration to ALS patients in Spain

ElevenLabs announced a partnership with the Spanish ALS association adELA to give ALS patients diagnosed in Spain free access to its voice restoration and preservation technology. This is the first project of the ElevenLabs Impact Program in Spain, and Spain is also the first country to offer this technology through a nationwide patient association. Patients whose voices have not yet changed noticeably can record new samples; those whose voices have already changed can upload old recordings to generate their own digital voice. It is estimated that there are about 4,000–4,500 ALS patients in Spain, many of whom will eventually lose the ability to speak.

View story:ElevenLabs partners with adELA to provide free voice restoration to ALS patients in Spain

Artificial Analysis evaluates Cartesia Managed Agents voice agent

Artificial Analysis evaluated Cartesia's fully managed voice agent solution, Cartesia Managed Agents, on the Speech Agent Arena. The solution achieved an Elo of 1115, 98 points higher than the previous Cartesia Line (1017), with task success rate rising from 77.5% to 91.7%. The platform runs in a cascade around a developer-selected LLM using Cartesia's in-house Ink speech-to-text and Sonic text-to-speech models; this evaluation used GPT-5.5 and the two voices Skylar and Daniel. On agent tasks requiring tool calls, the Elo was 1112. There are currently no further developments.

Artificial Analysis evaluates Cartesia Managed Agents voice Agent

Artificial Analysis evaluated Cartesia's fully managed voice agent solution, Cartesia Managed Agents, on the Speech Agent Arena. The solution achieved an Elo of 1115, 98 points higher than the previous Cartesia Line (1017), with task success rate rising from 77.5% to 91.7%. The platform runs in a cascade around a developer-selected LLM using Cartesia's in-house Ink speech-to-text and Sonic text-to-speech models; this evaluation used GPT-5.5 and the two voices Skylar and Daniel. On agent tasks requiring tool calls, the Elo was 1112. There are currently no further developments.

View story:Artificial Analysis evaluates Cartesia Managed Agents voice agent

ElevenLabs Conducts Shopper Research on Retail AI Voice Agents

ElevenLabs conducted three studies with shoppers in the UK, US and several other countries to examine shoppers' preferences for interacting with retail AI voice agents. The research shows that when shoppers were first asked to actually use an ElevenLabs voice agent before answering, nearly half chose "speak and listen to a reply," about 30% chose "speak and read text on screen," and voice ranked first across all six shopping tasks; among respondents who had not tried it, typing preferences led voice by more than 4:1. The research also found that the more serious the problem, the less willing shoppers were to hand it over to an agent, with only 34% choosing an agent in an erroneous charge scenario; if an agent fails once, 61% want to be transferred to a human.

Block releases full-duplex speech model Dessa

Block's AI research lab has released the full-duplex speech model Dessa, opening a waitlist as a research preview. According to MarkTechPost, the model was trained only on human voice conversation audio, without using text or pretrained models, and can run locally on a MacBook Pro M4 Max or higher configuration, faster than real time; training used one NVIDIA GB200 NVL72 (72 GPUs). No benchmark scores, parameter count, weights, paper, or API have been released so far.

View story:Block releases full-duplex speech model Dessa

Databricks launches voice Agent solution and open-sources DIVA blueprint

On October 29, 2026, Databricks launched a voice agent solution and open-sourced the DIVA blueprint as the starting point for the next scenario. The solution emphasizes that the value of voice agents lies not in being able to converse, but in combining context, ontology, and governed data to understand callers in real time and execute rules. Its prototype uses LiveKit to carry calls, Deepgram for speech-to-text and default voice, and ElevenLabs to provide expressive voice, and can hand calls over to a human. The agent runs where the customer's data resides, connecting to order, loyalty, and CRM systems through read-only queries. Currently, the solution uses the DIVA blueprint as the starting point for subsequent scenarios.

View story:Databricks launches voice Agent solution and open-sources DIVA blueprint

ElevenLabs conducts research on shoppers using retail AI voice agents

ElevenLabs conducted three studies with shoppers in the UK, US and several other countries to examine shoppers' preferences for interacting with retail AI voice agents. The research shows that when shoppers were first asked to actually use an ElevenLabs voice agent before answering, nearly half chose "speak and listen to a reply," about 30% chose "speak and read text on screen," and voice ranked first across all six shopping tasks; among respondents who had not tried it, typing preferences led voice by more than 4:1. The research also found that the more serious the problem, the less willing shoppers were to hand it over to an agent, with only 34% choosing an agent in an erroneous charge scenario; if an agent fails once, 61% want to be transferred to a human.

View story:ElevenLabs Conducts Shopper Research on Retail AI Voice Agents

ElevenLabs research claims voice outperforms text chat in complex conversations

On October 27, 2026, ElevenLabs released research findings stating that in complex conversations, voice achieves a higher success rate than text chat. After comparing hundreds of production-grade agents running both voice and chat simultaneously, the study found that in complex conversations lasting more than two minutes, voice had a success rate 32% higher than text chat, and resolved issues about one-third faster when successful. In a simulated flight cancellation scenario, 85% of participants preferred voice. The research also showed that an average end-to-end latency reduction of 16% corresponded to a 26% increase in success rate, and the proportion of calls hung up within 15 seconds dropped from 21.1% to 3.8%. Currently, this conclusion is published by a single source, the ElevenLabs Blog, and has not yet been cross-verified by other independent sources.

View story:ElevenLabs study says voice outperforms text chat in complex conversations

ElevenLabs launches synthetic voice detection in ElevenAgents

On October 27, 2026, ElevenLabs launched a synthetic voice detection feature in ElevenAgents that can determine in real time whether a caller is human or AI. The feature supports prioritizing human callers in the queue, routing AI calls to dedicated agents for handling, and blocking sensitive operations such as password resets. An article published the same day on the ElevenLabs Blog also noted that researchers have tracked 84 cases of surges in public service applications, appeals, and complaints across 11 countries, calling it "agentic flooding"; the Czech Republic's national welfare and employment hotline uses AI agents to handle about 5,000 calls per day, independently resolving about 85% of calls, with its answer rate rising from about 60% to near full coverage. There is currently no further progress following the launch of this feature.

View story:ElevenLabs launches synthetic voice detection in ElevenAgents

ElevenLabs study says voice outperforms text chat in complex conversations

On October 27, 2026, ElevenLabs released research findings stating that in complex conversations, voice achieves a higher success rate than text chat. After comparing hundreds of production-grade agents running both voice and chat simultaneously, the study found that in complex conversations lasting more than two minutes, voice had a success rate 32% higher than text chat, and resolved issues about one-third faster when successful. In a simulated flight cancellation scenario, 85% of participants preferred voice. The research also showed that an average end-to-end latency reduction of 16% corresponded to a 26% increase in success rate, and the proportion of calls hung up within 15 seconds dropped from 21.1% to 3.8%. Currently, this conclusion is published by a single source, the ElevenLabs Blog, and has not yet been cross-verified by other independent sources.

ElevenLabs launches synthetic voice detection in ElevenAgents

On October 27, 2026, ElevenLabs launched a synthetic voice detection feature in ElevenAgents that can determine in real time whether a caller is human or AI. The feature supports prioritizing human callers in the queue, routing AI calls to dedicated agents for handling, and blocking sensitive operations such as password resets. An article published the same day on the ElevenLabs Blog also noted that researchers have tracked 84 cases of surges in public service applications, appeals, and complaints across 11 countries, calling it "agentic flooding"; the Czech Republic's national welfare and employment hotline uses AI agents to handle about 5,000 calls per day, independently resolving about 85% of calls, with its answer rate rising from about 60% to near full coverage. There is currently no further progress following the launch of this feature.

Ollama launches speech model Phonon, supporting text-to-speech

Ollama launched its speech model Phonon on October 24, 2026, beginning support for text-to-speech. According to the Ollama Blog, Phonon is a series of small text-to-speech models from Gradium, currently available on macOS, with Windows and Linux support coming soon. On a MacBook Pro with M5 Max, Ollama uses MLX acceleration to run 35% faster than Gradium's own runtime, with word error rate on par with the Hugging Face Transformers reference implementation. Phonon is compatible with OpenAI's /v1/audio/speech interface, so existing OpenAI SDK code only needs its base URL changed. On licensing, phonon is released under Apache 2.0, while the 40M model uses the CC BY-NC 4.0 license and is restricted to non-commercial use. There are currently no subsequent developments.

Ollama launches speech model Phonon

Ollama launched its speech model Phonon on October 24, 2026, beginning support for text-to-speech. According to the Ollama Blog, Phonon is a series of small text-to-speech models from Gradium, currently available on macOS, with Windows and Linux support coming soon. On a MacBook Pro with M5 Max, Ollama uses MLX acceleration to run 35% faster than Gradium's own runtime, with word error rate on par with the Hugging Face Transformers reference implementation. Phonon is compatible with OpenAI's /v1/audio/speech interface, so existing OpenAI SDK code only needs its base URL changed. On licensing, phonon is released under Apache 2.0, while the 40M model uses the CC BY-NC 4.0 license and is restricted to non-commercial use. There are currently no subsequent developments.

View story:Ollama launches speech model Phonon, supporting text-to-speech

Block releases full-duplex speech model Dessa

Block's internal AI research lab released the full-duplex speech-to-speech model Dessa on October 23. According to Unite.AI, the model was trained entirely from scratch on human conversational speech, can listen and speak simultaneously, and natively perceives emotion, sarcasm, and laughter; its inference is faster than real time and it can run locally on a MacBook Pro M4 Max or higher configuration, with the team completing training on a single NVIDIA GB200 NVL72. Block has released demo recordings and a public waitlist, and has opened early access to a small number of AI safety researchers. Other reports note that the model does not rely on a text intermediary and supports full-duplex conversation with the ability to interrupt and interject at any time, and that training consumes hundreds of times fewer tokens than large language models, but the author points out that this is currently only a small-scale research preview from Block's internal lab, the demos were selected by the company, and questions remain about how pure audio models perform on complex multi-step logical reasoning.