Whisper
SoftwareWikipedia
129
MENTIONS
24
EPISODES
18
PODCASTS
Search complete. 129 mentions across 24 episodes found for "Whisper".
Sep 15, 2026
Adam SchwabGUEST
But the manual stuff actually, and getting the data right in the right format and fixing the mistakes takes longer than actual prompting, bizarrely.
Lisa TehHOST
I need to get you onto Whisper, remember? I'm always like, Adam, use Whisper.
Lisa TehHOST
Especially because, yeah, listen to how fast you talk.
Lisa TehHOST
I mean, you talk, you'll be able to like, it might be like, cannot compute talking way too fast.
Pavankumar Reddy MuddireddyGUEST
So we have the trunk, which is a three B, uh, text model that we train, uh, that we call Mistral series of models, and audio input is provided to the model through a audio encoder.
Pavankumar Reddy MuddireddyGUEST
But unlike, say, Whisper or models like that, where the audio input goes through an encoder, which is then fed into the decoder through a cross attention, here audio, uh, encoder produces, uh, s- tokens, um, in this case, uh, embed continuous representations through embeddings.
Pavankumar Reddy MuddireddyGUEST
And then they're fed into the main decoder model, uh, just as a direct token input, similar to how you would feed text input.
Pavankumar Reddy MuddireddyGUEST
Uh, in, in the text case, it's a rather simple encoding scheme.
Pavankumar Reddy MuddireddyGUEST
You, uh, s- uh, you send it through a tokenizer and you get token IDs, and then you just have a embedding table.
Pavankumar Reddy MuddireddyGUEST
Uh, in this case, the encoder is a little bit more sophisticated.
Pavankumar Reddy MuddireddyGUEST
Uh, at least in the VoxelChat, the encoder is very close to Whisper encoder, although for the l-later models, we optimized it and adapted it.
Pavankumar Reddy MuddireddyGUEST
Uh, we just tried to reduce the number of layers to the, uh, minimum number required for getting the performance.
Thomas SteinerGUEST
One of the reasons we have the imperative API as well is that you can do negotiating.
Thomas SteinerGUEST
So let's say you built a transcription app and you can work well with Whisper Tiny, like the smallest model in the Whisper family.
Thomas SteinerGUEST
But then of course, if the user already locally has Whisper massive, whatever, giant in their cost cache, you wouldn't say no, right? You wouldn't download a tiny model if they already have the big one.
Thomas SteinerGUEST
So that's also some legit cases where you would just want to allow probing and just say like, yeah, this is a pretty good use case where we say, yeah, I know that I work with the smallest model, but if I have a bigger model already available, I will just take that.
Pete WardenGUEST
they're actually often pretty good.
Pete WardenGUEST
Like Quen, 1.7 billion, I think it is, actually does a better job than Whisper, significantly better job on some languages like Mandarin.
Pete WardenGUEST
And that is a, and, you know, the same goes for Gemma.
Pete WardenGUEST
The trouble is that as general purpose models, like one of the failure cases with Gemma is that you upload audio of some speech.
JamieHOST
In local model developments, Desert Ant Labs debuted 18 on-device models covering transcription, audio cleanup, personal data redaction, language identification and video clipping, built to run entirely on-device with no cloud roundtrip and no token costs, free up to 100,000 monthly active devices.
JamieHOST
Its transcription model turns 10 minutes of audio into text in two seconds on an iPhone, nearly five times faster than Whisper, and its redaction model catches most sensitive strings in a 12-megabyte footprint.
JamieHOST
These are exactly the utility features consumer apps default to a cloud model call for today, and a free, faster, offline alternative quietly erodes a slice of per-token API revenue.
JamieHOST
DeepSeek released v4, OneFlash, a 552 billion parameter multimodal model under an MIT license, claiming near parity with its own larger v4 Pro base while activating only a quarter of the parameters and cutting per-token cache memory to under a kilobyte.
Leo LaporteHOST
The first thing I noticed immediately is the dictation, thankfully, is decent.
Leo LaporteHOST
It's still not as good as the state-of-the-art, the Whisper dictation, for instance, that, uh, Paris uses.
Leo LaporteHOST
But it's, but it's a lot better.
Paris MartineauHOST
Whisper dictation is so good.
Matthew CassinelliGUEST
Oh, yeah.
Leo LaporteHOST
It's amazing.
Matthew CassinelliGUEST
I use stuff like Monologue on desktop.
Leo LaporteHOST
And because Apple's such a walled garden, uh, Apple's such a walled garden, you can't...
Leo LaporteHOST
The first thing I noticed immediately is the dictation thankfully is decent.
Leo LaporteHOST
It's still not as good as the state-of-the-art, the Whisper dictation, for instance, that, uh, Paris uses, but it's, but it's a lot better.
Paris MartineauHOST
Whisper dictation is so good.
Matthew CassinelliGUEST
Oh, yeah.
Leo LaporteHOST
It's amazing.
Matthew CassinelliGUEST
... stuff like Monologue on the desktop
Leo LaporteHOST
... such a walled garden, uh, Apple's such a walled garden, you can't...
Leo LaporteHOST
I mean, on all, on my computers I can use Whisper AI, but I, I, I can't really use it on the iPhone.
![[YOU] on AI · Ahora](https://particle.news/cdn-cgi/image/format=auto,width=128/https://cdn.particle.pro/url/media/4e91d5b0-2117-58a6-bbc5-99f50507c704/4ba59d61cb1532c3192cbd071151c34c20f282a4cc8dbcff1e4711bdb0859bab)
Edo SegalHOST
En el frente del software empresarial, Daniel tiene dos lanzamientos que llegaron de madrugada.
DanielCORRESPONDENT
Microsoft lanzó en silencio MI Transcribe 2 a diez centavos por hora, un movimiento de presión directa sobre los precios de Whisper y Assembly AI.
DanielCORRESPONDENT
Y xAI abrió Grok Bot a equipos empresariales, añadiendo controles de política y barreras de seguridad para despliegue autónomo.
DanielCORRESPONDENT
El primer nivel formal de producto empresarial de xAI.
DanielCORRESPONDENT
Ahora cubre sesenta idiomas, frente a cuarenta y tres en junio, y añade diarización de hablantes, estilos configurables y marcas de tiempo a nivel de palabra.
DanielCORRESPONDENT
Microsoft afirma que ocupa el primer lugar en el benchmark Fluers, ejerciendo presión directa sobre OpenAI Whisper, Google y Eleven Labs en el mercado de transcripción empresarial.
Edo SegalHOST
Cinco historias, un hilo conductor.
Edo SegalHOST
La infraestructura de la IA diplomática, de hardware, de modelos y de precios está siendo renegociada en cada capa de manera simultánea.
Laura OsborneGUEST
No, I think it's been a lot of conversations at the moment have been really purpose driven.
Laura OsborneGUEST
So it's things like Whisper comes up a lot because people are specifically looking for transcription and translation and those kinds of storylines, which is, you know, if you... search for what do I use with transcription? What's a good model? Wisp is obviously going to be the first one that comes up.
Laura OsborneGUEST
So I do think there's still a bit more of an education piece to be done around this.
Laura OsborneGUEST
And I think it's evolved a lot.
14 more episodes mention Whisper.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.