| 摘要: | 「說話」是一個極其複雜的動作，除了需要構音器官彼此協調運作之外，大腦神經訊息的傳遞和對構音器官的控制也格外重要，隨著功能性磁振造影 (Functional Magnetic Resonance Imaging, fMRI)、擴散張量影像 (Diffusion Tensor Imaging, DTI) 等技術的成熟，大腦與說話的神經關聯性逐漸被揭開，除了皮質區域的控制功能，皮質下組織對說話的影響也開始受到關注和重視，然而針對基底神經核 (Basal ganglia) 的研究仍存在著限制，此外許多言語障礙也沒有找到有效的治療方式。因此，本研究的目的在研究基底神經核和說話之間的關聯性，藉由數學計算模型模擬基底神經核參與說話運動的過程，並從異常狀態下的模擬結果來預測可能發生的言語障礙現象，最後結合本實驗室先前研究的構音模型，找出言語治療的有效方法。
本研究使用以類神經網路為基礎之大腦訊號模型-GODIVA (Gradient Order Direction Into Velocities Articulator)，該模型模擬左下額葉溝 (Left inferior frontal sulcus)、前運動輔助區 (Pre-supplementary motor area)、額葉島蓋 (Frontal operculum) 以及尾狀核 (Caudate) 迴路，來產生說話所需的大腦訊號。而實驗方法為延伸該模型加入運動輔助區(Supplementary motor area)、殼核 (Putamen) 迴路等新的區塊至模型中，再模擬三種情形下大腦產生說話的過程，並與原版模型做比較。在正常情形部分，與原版相比第一音節會因為經過尾狀核迴路而增加約100毫秒的延遲時間，不過能提升模擬結果的準確度，顯示基底神經核對序列性的說話任務具有控制能力。在異常情形下，多巴胺 (Dopamine) 濃度的異常會影響多巴胺受體第一型 (D1) 和第二型 (D2) 的活動程度，第一型會導致所有音節的延遲，然而不同的音節會有不同的延遲程度；第二型會使刺激音節轉移的強度降低，導致音節轉移的時間點向後延遲；在白質纖維 (White matter fiber) 受損情形下，也會導致刺激音節轉移的強度降低，出現音節無法轉移的現象。上述異常情形皆表示基底神經核的受損會導致說話運動障礙，然而造成障礙的主因並非皆由單一因素所造成，也可能為多重原因影響之下所導致的結果。
最後，將修改後的GODIVA模型，與本實驗室先前開發具有中文聲調之構音模型-DIVA模型結合，達到訊號由大腦下達指令，構音器模擬字詞之功能，透過真實發聲探討實際言語障礙與大腦神經的關聯性，然而受限於GODIVA模型考慮的控制參數，以及DIVA模型發聲構造不夠完善，只能做到模擬母音以及部分子音，和產生類似口吃重複發聲的現象。未來希望能加入更多控制因素，包含加入尾狀核迴路之間接路徑，模擬產生失語症等言語障礙，以及將DIVA聲道模型切割的更細緻，達到構音器精準模擬所有發音。
;“Speech production” is an extremely complex action. In addition to the coordination of articulators, the transmission of neural signals and control of articulators are also important. With the advent of functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI), the neural correlation between brain and speech has gradually been uncovered. Apart from the control function of cortical areas, the influence of subcortical areas on speech has received more and more attention. However, the research methods on the basal ganglia (BG) is still limited, and there are no effective treatments for many speech disorders so far. Therefore, the purpose of this study is to investigate the correlation between basal ganglia and speech production. By simulating the process of the basal ganglia participating in speech production with a computational model, we predict the possible motor speech disorders from simulation results under abnormal conditions. Finally, we combined our model with the articulation model previously studied in our laboratory to find out effective ways for speech therapy.
GODIVA (Gradient Order Direction Into Velocities Articulator), a neural network based model, simulated the left inferior frontal sulcus, the pre-supplementary motor area, the frontal operculum and the caudate nucleus circuits to generate brain signals used in speech production in this study. Our approach was to add new blocks such as the supplementary motor area and the putamen nucleus circuits to the model and then simulate the process of speech production in three different conditions. After that, we compared simulation results with the original GODIVA model. In normal condition, even though passing through caudate nucleus circuit causes a delay with the first syllable at about 100 milliseconds compared to the original GODIVA model, the accuracy of the simulation results could be significantly improved. It suggests that the basal ganglia have control ability for sequential speech tasks. In another condition, abnormal level of dopamine would affect the activity of dopamine receptor D1 and D2. D1 would cause delays in all syllables, but the delays of syllables would be different. D2 would decrease the intensity of the syllable switching activity, causing delays to the switch point of syllable. In the last condition, white matter fiber impairments would also reduce the intensity of the activity so that the syllable could not be changed to the next one. The above-mentioned abnormal conditions suggest that the impairment of basal ganglia would lead to motor speech disorders. The major source of the disorder is not always originated from a single factor, but may also from the results of multiple factors.
Finally, we combined the GODIVA model with the DIVA model previously developed in our laboratory with Chinese tones to build the function that the GODIVA model transmits brain signal instruction to control DIVA model for producing speech sound. In this way, we could find out the correlation between brain and motor speech disorder. Unfortunately, due to the limit of number of control parameters used in the GODIVA model and integrity of vocal tract used in DIVA model, the model only can simulate vowels and some consonants, and produce stuttered-like repeated sounds. In future studies, we hope more factors could be considered such as the indirect pathway of caudate nucleus circuit, so the model could simulate other motor speech disorder such as aphasia. Besides, the vocal tract shape of the DIVA model could be modified to accurately simulate all syllables. |
