Crate aurum_core 

aurum-core

Reusable on-device speech I/O library (experimental API).

• STT — local whisper.cpp by default; optional OpenRouter
• TTS — local ONNX KittenTTS (cargo feature tts, default on)
• Cleanup — rules or optional LLM post-edit

Tagline: Speech both ways. On-device by default.

The API may change without notice until a stable 0.1.0.

STT example

use aurum_core::audio::{AudioInput, WHISPER_SAMPLE_RATE};
use aurum_core::pcm::PcmBuffer;
use aurum_core::providers::{LocalWhisperProvider, TranscriptionOptions};
use std::path::PathBuf;

let provider = LocalWhisperProvider::new(PathBuf::from("/tmp/aurum-cache"))
    .with_progress(false)
    .with_local_only(false);
provider.preload("tiny-q5_1").await?;

let mut buf = PcmBuffer::dictation();
buf.push(&[0.0f32; 1600])?;
let result = provider
    .transcribe_pcm(
        buf.samples(),
        &TranscriptionOptions {
            model: "tiny-q5_1".into(),
            language: "en".into(),
            timestamps: false,
            cancel: None,
        },
    )
    .await?;
let _ = AudioInput::from_pcm_slice(buf.samples(), WHISPER_SAMPLE_RATE)?;
println!("{}", result.text);
aurum_core::providers::local::clear_context_cache();

Re-exports

pub use audio::load_audio;

pub use audio::AudioInput;

pub use audio::WHISPER_SAMPLE_RATE;

pub use cancel::CancelFlag;

pub use cleanup::apply_cleanup;

pub use cleanup::apply_cleanup_with_segments;

pub use cleanup::cleanup_text;

pub use cleanup::CleanupProviderKind;

pub use cleanup::CleanupResult;

pub use cleanup::CleanupStyle;

pub use cleanup::OpenRouterCleanup;

pub use cleanup::RulesCleanup;

pub use cleanup::SegmentCleanupPolicy;

pub use cleanup::TextCleanup;

pub use config::Config;

pub use error::Result;

pub use error::TranscriptionError;

pub use model::list_models;

pub use model::DownloadProgress;

pub use model::EnsureModelOptions;

pub use model::ModelInfo;

pub use model::ModelStatus;

pub use output::format_result;

pub use output::OutputFormat;

pub use pcm::PcmBuffer;

pub use providers::LocalWhisperProvider;

pub use providers::OpenRouterProvider;

pub use providers::Segment;

pub use providers::TranscriptionOptions;

pub use providers::TranscriptionProvider;

pub use providers::TranscriptionResult;

pub use window::PartialClock;

pub use window::PartialWindowPolicy;

pub use tts::format_model_list as format_tts_model_list;

pub use tts::format_voice_list as format_tts_voice_list;

pub use tts::list_models as list_tts_models;

pub use tts::list_voices as list_tts_voices;

pub use tts::write_wav_i16_mono_atomic;

pub use tts::BackendKind as TtsBackendKind;

pub use tts::LocalTtsProvider;

pub use tts::SynthesisOptions;

pub use tts::SynthesisProvider;

pub use tts::SynthesisResult;

pub use tts::DEFAULT_TTS_MODEL;

pub use tts::DEFAULT_TTS_VOICE;

Modules

audio

Audio loading and conversion.

cancel

Cooperative cancellation for long-running local transcription.

cleanup

Post-transcription text cleanup ( “flow”).

config

Configuration loading for Aurum.

error

Error taxonomy for Aurum.

model

Local whisper.cpp model management: resolve, download, cache, list.

output

Output formatters: txt, srt, json.

pcm

In-memory PCM buffer for mic / streaming hosts (no files, no ffmpeg).

postprocess

Shared transcript cleanup applied after every provider.

providers

Transcription provider abstraction.

tts

Local text-to-speech (TTS) — ONNX KittenTTS + MIT G2P.

window

Rolling-window helpers for host-driven “partial-like” decode loops.