AI Voice and Speech Release Notes
Release notes for AI voice synthesis, text-to-speech and audio generation tools
Products (13)
Latest AI Voice and Speech Updates
- Oct 17, 2026
- Date parsed from source:Oct 17, 2026
- First seen by Releasebot:Oct 17, 2026
API: regional hosts, api.speechify.ai routes speech by location, and api.sws.speechify.com stops serving on 19 January 2027
Speechify adds regional API hosts for Europe and US East, routing synthesis to the nearest region by default while keeping existing keys and requests compatible. It also sets a 2027 retirement date for the legacy API host and directs users to api.speechify.ai.
eu.api.speechify.ai and us.api.speechify.ai pin the region that runs synthesis
Two new hosts serve the text-to-speech and voice endpoints from one region: eu.api.speechify.ai from Europe and us.api.speechify.ai from US East. api.speechify.ai is unchanged and stays the default.
Only the base URL changes. The same API key, requests and responses work on every host.
A model with no capacity in the region still runs. The request goes to the nearest pool that serves the model, usually US East.
Not data residency. Your account, keys, voices, usage and logs stay in the United States on every host.
Rate and concurrency limits are counted per region. Spend limits, project budgets and your credit balance count spend from every region.
No API version change is needed. See Regions.
api.speechify.ai serves speech from the region nearest the caller
Text-to-speech and voice requests to api.speechify.ai now run in the region nearest the caller: Europe (the Netherlands) for callers closer to it, US East for everyone else.
- Every request, from today. Speech and voices moved in steps on 17 October, and streaming (/v1/audio/stream, /v1/audio/stream/with-timestamps and /v1/audio/stream/ws) followed that evening.
- All of a caller’s synthesis now runs in one region. A stream from Europe runs on the model’s capacity in Europe where it has some.
- Nothing to change. The same API key, requests and responses work. To keep one region, use us.api.speechify.ai or eu.api.speechify.ai.
- Rate and concurrency limits are counted in the region that serves each request. Spend limits, project budgets and your credit balance count spend from every region.
- Not data residency. Your account, keys, voices, usage and logs stay in the United States.
api.sws.speechify.com stops serving on 19 January 2027
api.speechify.ai is the only production host for the SpeechifyAI API. The legacy host api.sws.speechify.com keeps answering until Tuesday 19 January 2027 at 14:00 UTC. From then, every request to it gets a 308 Permanent Redirect to the same path on api.speechify.ai. This replaces the 19 February 2026 entry, which said the legacy host was not being deprecated.
- Change your base URL to https://api.speechify.ai. The same API keys, requests and responses work there.
- Using an official SDK? Upgrade it or remove the hard-coded base URL. Our SDKs have defaulted to api.speechify.ai since February 2026.
- An integration still on the legacy host stops working on that day. Most HTTP clients either do not follow a redirect on a POST or drop the Authorization header when the redirect goes to another host.
- Responses from the legacy host carry Deprecation and Sunset headers with the date, so an API gateway or SDK that reads them can warn you.
Workspaces that called the legacy host in the last 30 days get an email, and a reminder 30 days before the date.
Original source - Oct 16, 2026
- Date parsed from source:Oct 16, 2026
- First seen by Releasebot:Oct 17, 2026
TTS language field
Fish Audio adds an optional language setting for TTS text normalization and word timestamps across API and WebSocket routes.
TTS language field
Text-to-speech requests accept an optional language field, a BCP-47 tag of the text such as en, es-MX, or zh-TW. It selects the rules used to spell out numbers, dates, units, and symbols before synthesis, and the language used for word timestamps. It does not change the voice, accent, or audio quality. Omit it to keep detecting the language per text chunk.
Available on POST /v1/tts, the with-timestamps routes, and the WebSocket start event. The official SDKs don't expose it yet.
Language | Supported normalization languages
Original source All of your release notes in one feed
Join Releasebot and get updates from Speechify and hundreds of other software products.
- Oct 15, 2026
- Date parsed from source:Oct 15, 2026
- First seen by Releasebot:Oct 17, 2026
October 15, 2026
Murf adds IPA and pronunciation dictionary controls to Falcon 2 across Stream Speech and WebSockets.
Pronunciation control for Falcon 2
Falcon 2 now gives you two ways to set exactly how a word is pronounced:
- IPA in your text: write the word in IPA between forward slashes inside double quotes, for example Welcome to "/mɜːrf/". Works on Stream Speech and WebSockets.
- Pronunciation dictionary: pass pronunciationDictionary on Stream Speech, the same parameter Synthesize Speech supports for Gen2. On WebSockets, send pronunciation_dictionary in voice_config once per context. Each entry sets a word's IPA or replacement words, for every time the word appears.
See Pronunciations.
Original source - Oct 14, 2026
- Date parsed from source:Oct 14, 2026
- First seen by Releasebot:Oct 14, 2026
Ask Plaud is about to work in a whole new way.
Plaud introduces Chat and Work in Ask Plaud: Chat answers questions using recordings and knowledge, while Work uses an AI agent to complete tasks, create documents, and send results through connected apps.
Ask Plaud: Chat and Work
Ask Plaud gets Chat and Work: two ways to get help from your recordings and knowledge base. Chat is for questions. Work is for getting things done.
Use Chat when you want an answer. Ask what was decided in last Tuesday's meeting, who owns an action item, or what your team discussed last month. You ask, Chat answers. You stay in the conversation the whole time.
Use Work when you want something done. Describe what you need: a follow-up email drafted, a summary posted to Slack, a brief ready before Monday. Plaud Agent takes it from there and hands you back the result.
- Your goal: Get an answer | Get something done
- The task: A question about your recordings | Multi-step: write, schedule, send, generate
- Your role: In the conversation throughout | Describe it once, review what comes back
- The output: An answer to read or act on | A finished file or a completed action
- Cost: Free | Uses Plaud Credits
What Chat does
Chat draws on your recordings, knowledge base, and memory to answer you directly. Reference a recording or an image, ask your question, and it gives you accurate replies. Connect Google Calendar and Chat can check your schedule too. It does not write files, trigger automations, or send anything on your behalf.
What Work adds
Work lets Plaud Agent act, not just respond. Your Skills define how tasks should run. Routines handle recurring work on a schedule. Artifacts turn conversations into real documents: PDFs, slide decks, Word files, ready to share. Connectors send the output straight to Slack, Notion, Linear, and more.
How it works in practice
Every new Ask Plaud conversation starts with a choice: Chat or Work. Pick one before your first message, and it stays for that conversation. If Chat cannot handle what you asked, it tells you and points you to Work.
Chat is the default. After your first conversation, Ask Plaud remembers which one you used last on Plaud mobile app, web, and desktop.
Original source - Oct 14, 2026
- Date parsed from source:Oct 14, 2026
- First seen by Releasebot:Oct 14, 2026
API: the 20,000-character limit on the stream endpoints counts spoken text, not SSML tags
Speechify updates its streaming audio API so SSML tags no longer count toward the 20,000-character spoken-text limit. Requests can include up to 32,000 characters including markup, with clearer limit-specific validation errors.
API: the 20,000-character limit on the stream endpoints counts spoken text, not SSML tags
POST /v1/audio/stream and POST /v1/audio/stream/with-timestamps now accept up to 20,000 characters of spoken text, the characters a request is billed for. SSML tags no longer count toward the limit, as on /v1/audio/speech since 9 Oct.
The number is unchanged; only what it counts is. Tags such as , , and are not counted, an entity such as & counts as one character, and a counts its alias.
The input as sent can be up to 32,000 characters, markup included.
A refusal says which limit it hit. It is still 400 validation_failed, and the message names either the spoken-text limit or the 32,000-character ceiling.
Nothing that was accepted before is refused. Plain text is counted exactly as before.
/v1/audio/speech is unchanged: 2,000 spoken characters, and up to 20,000 as sent.
No API version change is needed. See API Limits.
Original source - Oct 14, 2026
- Date parsed from source:Oct 14, 2026
- First seen by Releasebot:Oct 14, 2026
Detection Learned to Reason. Here's What That Meant for Q3.
Resemble expands its AI detection platform with DETECT-World for deepfake detection, Investigator for automated forensic analysis, and Agent Detection to distinguish human, bot, and AI traffic. The updates bring stronger detection and clearer evidence to decisions across media and online activity.
In September, our CTO Tedi Papajorgji took the stage at the Exploratorium in San Francisco for GMI's Summer Signal conference.
More than 200 builders, investors, and AI founders were in the room and teams from Runway, Luma, MiniMax, and Inworld AI were showing what’s possible with real-time video generation, agentic filmmaking, and interactive AI characters.
Tedi opened his presentation with a guessing game. The audience saw a video of a press conference, a voice clip, and a photo of a dog on a couch, then had to decide whether each was real or fake.
All three were synthetic and I would say less than 10% of the folks in the room guessed all three correctly - and that was after being told they needed to decide if the content was real or fake - they were already on high alert.
Everyone else focused on what generative AI can create (and how realistic it is) and we wanted to bring home the fact that the best human detection is wrong, most of the time.
That became a pretty good description of our work in Q3. After officially sunsetting voice generation as a paid product, we put the full company behind detection. Generative models kept changing, and our work increasingly focused on detection that could understand more than the fingerprints of models it had already seen.
DETECT-World: Looking Beyond the Fingerprints
Our previous-generation model, DETECT-3B Omni, was already strong, ranking first on DFBench and covering audio, image, and video across 51 languages. But like most detection models, it relied heavily on learning the artifacts that known generators leave behind and looking for those artifacts in new content.
That gets harder as the number of generators grows. A new model launches, your detector has never seen its outputs, and performance can drop until you retrain it. You end up constantly collecting new examples and teaching the model what to look for.
DETECT-World takes another approach where it still looks for the artifacts associated with generated content, but it also looks at whether the content itself makes sense.
Is the lighting consistent across a face? Do the shadows fall the way they should based on the light source? Does movement in the background make sense from one frame to the next? Do the lips actually match the audio?
Those signals are useful because they aren't tied to one particular generator and the physics don't change when a new generator launches.
The results for our new model have also been really compelling. Video deepfake detection accuracy increased from 80.96% to 98.2%, while image accuracy went from 86.70% to 95.8%. Audio accuracy reached 99.47%, with a false positive rate of 0.66%, and latency dropped 66% to 399 milliseconds.
On the independent Podonos audio deepfake detection benchmark, DETECT-World ranked first out of 18 evaluated systems as of August 2026. It was the only system on the board with both false positive and false negative rates below 1%. Our previous model, DETECT-3B Omni, ranked second.
We also tested DETECT-World against Haotian AI's real-time face-swap tool, which was new to the model. It identified the generated content with approximately 95% accuracy without prior exposure to the generator, retraining, or fine-tuning.
That idea of looking beyond known patterns also shaped how we thought about the rest of the detection stack this quarter.
At GMI, we showed another side of that work with Model Behavior, an exhibit built from UMAP visualizations our models produce internally. Each dot represents a piece of media, with similar and different examples arranged into clusters.
Seen together, the visualizations start to look almost organic: a flock of birds, a river delta, shapes forming and separating as the model finds relationships in the data.
For us, it was a way to make something that normally happens inside a model visible to people. Detection comes down to understanding how media fits with what the model has learned about real and manipulated content, rather than relying on a single artifact.
Fully Automated, Forensic-grade Investigation
That same idea came up when we started working on what happens after a detection result.
A trust and safety analyst looking at a piece of social media content with a 94% manipulation probability still needs to know what kind of manipulation occurred, how strong the evidence is across different forensic dimensions, and whether the content actually needs to be removed.
That's why we launched Investigator this quarter: six managed forensic investigators available through our API, each built around a specific verification workflow.
They cover social media moderation, insurance claim materials, breaking news footage, documents and receipts, forensic evidence, and identity documents and presented individuals.
Each investigation brings together multiple forensic analysis layers and returns evidence and explanations alongside the result. The goal is to give the person reviewing the content enough context to make a decision without having to piece the investigation together manually.
This came up repeatedly in conversations with enterprise customers. The detection itself was highly accurate, but when next steps require more than “is this AI or not,” the Investigator’s full context, multi-API results and recommendations were needed..
Detecting What Entity is Operating - Bot, AI or Human?
For the past three years, when we talked about detection, we were usually talking about media: is this audio, image, or video AI-generated?
Now there's another question worth asking: what if the thing interacting with your website or application is AI?
Agent Detection is our System One model and a new product for us. Website and app owners can add a single line of code and start getting visibility into their traffic within minutes.
Instead of relying on user-agent strings, which agents can spoof, the model looks at behavioral signals such as cursor movement, typing rhythm, and interaction patterns. Each verdict includes reasoning, so operators can see why a visitor was classified as a human, an AI agent, or something that needs further review.
That gives organizations choices about what to do with different types of traffic. They might reroute agents to APIs or MCP servers, restrict access to sensitive areas such as billing or credential management, or prevent specific activities such as account deletion.
I don't think the answer is to block every AI agent. Some are doing useful things, and personal usage is already exploding, so organizations will have to set policy for routing, allowance and restriction. The problem is knowing what is happening in your systems and having enough information to decide how to respond, and solving it starts by knowing whether you are dealing with a human, a bot or an agent..
In that sense, Agent Detection is a natural extension of the work we've already been doing. We started by asking whether a piece of media was generated or manipulated in both content and calls and now we're also looking at behavior, whether the thing interacting with a system is an AI agent and what it is doing.
Seeing the Work in the Real World
Some of the recognition we received this quarter reflected that broader approach.
DETECT-World won Best Technological Solution at the Global Anti-Scam Summit America and at Visual 1st, we won Best Technology, presented by Google's Navin Sarma.
The awards were encouraging, but I found the conversations around them more interesting. People are starting to ask questions that go beyond “is this fake?” They want to know where something came from, how confident they should be in the result, and what they can actually do with that information.
That's also where watermarking fits into the picture. Detection can identify signs of manipulation, while watermarking can help establish provenance.
We have worked on both for years, and signing the EU Code of Practice on Transparency of AI Generated Content alongside Google, Anthropic, Meta and more as a technology provider of marking and detection solutions was an important step in making that part of our approach more explicit.
Zohaib also joined a panel at Google's AI safety event in September on agentic safety and observability. A point he returned to was that organizations need visibility into what is happening before they can put meaningful controls around it.
That applies to new AI systems, but it also shows up in much older infrastructure. We're seeing it in legacy call center systems dealing with voice-clone calls, just as we're seeing it in websites dealing with AI agents.
What We Owe Each Other
All of this brings me back to that room at the Exploratorium.
The audience had to decide whether a few pieces of media were real or fake. A year ago, that might have felt like a reasonable question for people to answer on their own but now it’s impossible.
As generative models get better at reproducing reality, our own judgment becomes less reliable. Sometimes we decide something is fake because it doesn't fit our expectations and other times, we believe something because it confirms what we already expect to be true.
Models can approach the problem differently. They can look across a much larger body of evidence and examine the relationships between things that are difficult for a person to see.
That's part of why we wanted Model Behavior at GMI. The visualizations gave people a way to see how a detection model organizes that evidence, rather than treating the final score as a black box.
The question underneath all of this is still a simple one: does this deserve your trust?
For us, answering that question increasingly means more than returning a score. It means understanding the content, investigating the evidence, knowing when something is new, and giving people enough context to decide what happens next.
What's Next
We're launching Resemble AI for Microsoft Teams shortly, bringing bot-free real-time deepfake detection into the meetings where teams are already making important decisions.
And that's just the start. We have a lot more new products and partnerships coming that will help organizations detect AI across content, calls, and behavior.
If any of this is relevant to what you are building or protecting, we would like to hear from you, book a call with our team.
Original source - Oct 14, 2026
- Date parsed from source:Oct 14, 2026
- First seen by Releasebot:Oct 14, 2026
Drama 3
Fish Audio releases Drama 3, a text-to-speech model directed in plain language for tone, pacing, and character, available via the API as drama-3.
Drama 3
Text-to-speech model you direct in plain language. Describe tone, pacing, and character without a fixed set of audio tags.
Use model ID drama-3 in the API.
Drama 3 guide | Text to Speech API
Original source - Oct 14, 2026
- Date parsed from source:Oct 14, 2026
- First seen by Releasebot:Oct 14, 2026
Fish Audio S1 retirement
Fish Audio is retiring S1 on December 31, 2026, routing its requests to S2.1-pro at the same price and recommending tag updates.
Fish Audio S1 retirement
S1 is deprecated and will be retired on December 31, 2026. After that date, requests that specify s1 are served by s2.1-pro and billed as s2.1-pro, at the same price.
To migrate, switch the model ID to s2.1-pro and convert (parenthesis) emotion tags to [bracket] tags.
Migrate from S1 | Model deprecations
Original source - October 2026
- No date parsed from source.
- First seen by Releasebot:Oct 14, 2026
15.6
MacWhisper adds Spanish, French, Japanese, Simplified Chinese, Swedish, and Finnish, plus automatic recording stop after meetings. It also improves translation feedback, dictation accuracy, and Qwen3-ASR transcription progress.
New
- Languages: MacWhisper is now available in Spanish, French, Japanese, Simplified Chinese, Swedish, and Finnish.
- Meetings: Enable Automatically Stop Recording to stop recording 10 seconds after a meeting ends, or dismiss the alert to keep recording.
Improvements
- Translation feedback: Submit translation feedback from the Help menu in every supported app language.
Bugfixes
- Fixed: Dictation dictionary entries now retain spaces, punctuation, numbers, and capitalization.
- Fixed: On-screen context no longer joins separate words together during dictation.
- Fixed: Qwen3-ASR transcription progress now updates correctly.
- October 2026
- No date parsed from source.
- First seen by Releasebot:Oct 14, 2026
15.0.2
MacWhisper improves waveform performance and fixes file dragging, transcript replacement, Obsidian exports, and macOS layout issues.
Improvements
- Recording: smoother waveform rendering and reduced processing overhead.
Bugfixes
- Dragging recordings from apps such as Voice Memos could get stuck, including in Batch and Podcast mode.
- Dropping a file onto the transcript editor could replace the displayed transcript with the file’s contents.
- Obsidian exports could fail when note titles contained slashes, colons, or other unsupported characters.
- Fixed a layout-related crash on macOS 27.
- Fixed the Roadmap layout and Manage Export Formats button styling in Settings.