Go On-Device

Argmax SDK

Start with Open-source. Scale with Pro.

WhisperKit _PRO

Ship frontier speech-to-text models on device with WhisperKit Pro

WhisperKit Pro is a turn-key SDK that brings frontier speech-to-text capabilities to your user's devices, faster and more accurate than most cloud APIs.
‍
Read our research paper.
Try it on our Playground app.

- import WhisperKit // Open Source WhisperKit
+ import Argmax // Pro Argmax SDK

+ await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

- let config = WhisperKitConfig(model: "large-v3-v20240930")
+ let config = WhisperKitProConfig(
+     model: "large-v3-v20240930",
+     modelRepo: "argmaxinc/whisperkit-pro"
+ )

- let whisperKit = try await WhisperKit(config)
+ let whisperKitPro = try await WhisperKitPro(config)

import Argmax

// Initialize Argmax SDK to enable Pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

let config = WhisperKitProConfig(
    model: "large-v3-v20240930",
    modelRepo: "argmaxinc/whisperkit-pro",
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/whisperkit-pro
)

let whisperKitPro = try await WhisperKitPro(config)
let transcript = try? await whisperKitPro.transcribe(audioPath: "path/to/audio.m4a").text

import Argmax

// Initialize Argmax SDK to enable Pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

let config = WhisperKitProConfig(
    model: "large-v3-v20240930",
    modelRepo: "argmaxinc/whisperkit-pro",
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/whisperkit-pro
)

let whisperKitPro = try await WhisperKitPro(config)

var transcription = "" // Confirmed transcription text
var hypothesisText = "" // Hypothesis text from most recent transcription
var latestAudioSampleIndex = 0 // Track the latest audio samples sent to the transcribe task

let transcribeTask = whisperKitPro.transcribeWhileRecording(
    audioCallback: {
        // Get latest audio samples
        let newAudioToTranscribe = yourRecordingAudio[latestAudioSampleIndex...]
        latestAudioSampleIndex = yourRecordingAudio.count - 1

        // Send the new audio samples to the transcribe task
        return AudioSamples(samples: newAudioToTranscribe)
    },
    resultCallback: { result in
        transcription += result.text
        hypothesisText = transcribeTask.hypothesisText

        // Let the transcribe task know it should continue
        return true
    }
)

import Argmax

// Initialize Argmax SDK to enable pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

let config = WhisperKitProConfig(
    model: "large-v3-v20240930",
    modelRepo: "argmaxinc/whisperkit-pro",
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/whisperkit-pro
)

let whisperKitPro = try await WhisperKitPro(config)
let options = DecodingOptions(wordTimestamps: true)
let words = try? await whisperKitPro.transcribe(audioPath: "path/to/audio.m4a", options: options).allWords

for word in words {
    print("Word: \(word.word)")
    print("Start time: \(word.start) seconds")
    print("End time: \(word.end) seconds")
    print("Probability: \(word.probability)")
}

import Argmax

// Initialize Argmax SDK to enable pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

let config = WhisperKitProConfig(
    model: "large-v3-v20240930",
    modelRepo: "argmaxinc/whisperkit-pro",
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/whisperkit-pro
)

let whisperKitPro = try await WhisperKitPro(config)

let options = DecodingOptions(wordTimestamps: true)
let result = try? await whisperKitPro.transcribe(audioPath: "path/to/audio.m4a", options: options)

// Export as SRT (substitute with VTT or JSON as needed)
let srtWriter = WriteSRT(outputDir: "path/to/output")
let srtResult = srtWriter.write(result: result, to: "transcript")

// Saved SRT Report: file:///path/to/output/transcript.srt

SpeakerKit _PRO

Ship state-of-the-art speaker recognition on device

SpeakerKit Pro is an evolution of Pyannote optimized for speed without compromising accuracy.

Read the research paper.
Try it on our demo App.

import Argmax

// Initialize Argmax SDK to enable Pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

// Initialize SpeakerKitPro with configuration
let config = SpeakerKitProConfig(
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/speakerkit-pro
)

// Create SpeakerKitPro instance
let speakerKitPro = try await SpeakerKitPro(config)

// Process audio for diarization
try await speakerKitPro.initializeDiarization(audioArray: audioFrames) { audioClip in
    Task {
        // Process each audio segment
        try await speakerKitPro.processSpeakerSegment(audioArray: audioClip)
    }
}

// Finalize processing and get diarization results
let diarizationResult: DiarizationResult = try await speakerKitPro.clusterSpeakers()

import Argmax

// Initialize Argmax SDK to enable pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

// Initialize SpeakerKitPro with configuration
let config = SpeakerKitProConfig(
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/speakerkit-pro
)

// Create SpeakerKitPro instance
let speakerKitPro = try await SpeakerKitPro(config)

let modelVAD = speakerKitPro.modelVAD()

// Predict voice activity true/false in ~17ms resolution
let vadResult: [Bool] = try await modelVAD.predict(audioArray: audioClip)

import Argmax

// Initialize Argmax SDK to enable pro access
await ArgmaxSDK.with(ArgmaxConfig(apiKey: "ax_*****"))

let config = WhisperKitProConfig(
    model: "large-v3-v20240930",
    modelRepo: "argmaxinc/whisperkit-pro",
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/whisperkit-pro
)

let whisperKitPro = try await WhisperKitPro(config)

// Initialize SpeakerKitPro with configuration
let config = SpeakerKitProConfig(
    modelToken: "hf_*****" // Request access at https://huggingface.co/argmaxinc/speakerkit-pro
)

// Create SpeakerKitPro instance
let speakerKitPro = try await SpeakerKitPro(config)

// Load the audio
let audioFrames = try AudioProcessor.loadAudioAsFloatArray(fromPath: "/path/to/audio.m4a")

// Process audio for diarization
try await speakerKitPro.initializeDiarization(audioArray: audioFrames) { audioClip in
    Task {
        // Process each audio segment
        try await speakerKitPro.processSpeakerSegment(audioArray: audioClip)
    }
}

let options = DecodingOptions(wordTimestamps: true)
let result = try? await whisperKitPro.transcribe(audioArray: audioFrames, options: options)

// Finalize processing and get diarization results
let diarizationResult: DiarizationResult = try await speakerKitPro.clusterSpeakers()

// If you have transcription results, you can combine them with diarization
let updatedSegmentsArray = diarizationResult.addSpeakerInfo(to: result)
for segments in updatedSegmentsArray {
    for segment in segments {
        for line in segment.formatWithSpeakers() {
            print(line)
            // Output:
            // Speaker 0 [0.70-3.08s]: Just set up speaker diarization with SpeakerKit Pro.
            // Speaker 1 [3.12-4.42s]: Is this really running on device?
            // Speaker 0 [4.94-6.10s]: Yes, it is and it's running super fast!
        }
    }
}

DiffusionKit

DiffusionKit Pro allows you to run the most advanced diffusion models on device.

We’ve converted the most advanced diffusion models available in order to run on Apple Silicon with CoreML and MLX. DiffusionKit Pro is currently undegoing development.

# Pipeline for Flux
from diffusionkit.mlx import FluxPipeline

pipeline = FluxPipeline(
  shift=1.0,
  model_version=argmaxinc/mlx-FLUX.1-dev,
  low_memory_mode=True,
  a16=True,
  w16=True,
)

# Image Generation
HEIGHT = 512
WIDTH = 512
NUM_STEPS = 4
CFG_WEIGHT = 0

image, _ = pipeline.generate_image(
  "a photo of a cat",
  cfg_weight=CFG_WEIGHT,
  num_steps=NUM_STEPS,
  latent_size=(HEIGHT // 8, WIDTH // 8),
)

COPY

Copied

# Pipeline for Flux
from diffusionkit.mlx import FluxPipeline

pipeline = FluxPipeline(
  shift=1.0,
  model_version=argmaxinc/mlx-FLUX.1-schnell,
  low_memory_mode=True,
  a16=True,
  w16=True,
)

# Image Generation
HEIGHT = 512
WIDTH = 512
NUM_STEPS = 4
CFG_WEIGHT = 0

image, _ = pipeline.generate_image(
  "a photo of a cat",
  cfg_weight=CFG_WEIGHT,
  num_steps=NUM_STEPS,
  latent_size=(HEIGHT // 8, WIDTH // 8),
)

COPY

Copied

# Pipeline for Stable Diffusion 3
from diffusionkit.mlx import DiffusionPipeline

pipeline = DiffusionPipeline(
	shift=3.0,
	use_t5=False,
	model_version=argmaxinc/mlx-stable-diffusion-3.5-large,
	low_memory_mode=True,
	a16=True,
	w16=True,
)     

# Image Generation
HEIGHT = 512
WIDTH = 512
NUM_STEPS = 50
CFG_WEIGHT = 5

image, _ = pipeline.generate_image(
  "a photo of a cat",
  cfg_weight=CFG_WEIGHT,
  num_steps=NUM_STEPS,
  latent_size=(HEIGHT // 8, WIDTH // 8),
)

COPY

Copied

# Pipeline for Stable Diffusion 3
from diffusionkit.mlx import DiffusionPipeline

pipeline = DiffusionPipeline(
	shift=3.0,
	use_t5=False,
	model_version=argmaxinc/coreml-stable-diffusion-3-medium,
	low_memory_mode=True,
	a16=True,
	w16=True,
)     

# Image Generation
HEIGHT = 512
WIDTH = 512
NUM_STEPS = 50
CFG_WEIGHT = 5

image, _ = pipeline.generate_image(
  "a photo of a cat",
  cfg_weight=CFG_WEIGHT,
  num_steps=NUM_STEPS,
  latent_size=(HEIGHT // 8, WIDTH // 8),
)

COPY

Copied

The On-Device AI Platform

On-Device AI

Why?

Go On-Device

Argmax SDK

WhisperKit _PRO

SpeakerKit _PRO

DiffusionKit

Subscribe for Updates

Articles

The On-Device AI Platform

On-Device AI

Why?

Go On-Device

Argmax SDK

WhisperKit PRO

SpeakerKit PRO

DiffusionKit

Subscribe for Updates

Articles

WhisperKit _PRO

SpeakerKit _PRO